Engineer-to-Engineer Note EE-326: Blackfin® Processor and SDRAM Technology

Engineer-to-Engineer Note EE-326
Technical notes on using Analog Devices DSPs, processors and development tools
a Visit our Web resources http://www.analog.com/ee-notes and http://www.analog.com/processors or e-
mail processor.support@analog.com or processor.tools.support@analog.com for technical support.
Blackfin® Processor and SDRAM Technology

Contributed by Fabian Plepp Rev 2 – December 11, 2008
Introduction
The Analog Devices Blackfin® family of processors provides an External Bus Interface Unit (EBIU) with
which to interface to SDRAM.
This EE-Note covers the following topics:
Register settings and their meaning
SDRAM initialization
SDRAM hardware design
Using less than 16MB of SDRAM
Performance optimization
Power optimization
This EE-Note discusses SDR-SDRAM (not DDR-SDRAM) devices only.
Furthermore, this document applies to ADSP-BF53x, ADSP-BF52x, ADSP-BF51x and
ADSP-BF561 processors only. It does not apply to ADSP-BF54x processors.
Although this document covers basic aspects of SDRAM functionality, you should read The ABCs of
SDRAM (EE-126)[1] for more information.
This document is separated into different topics, so you do not need to read the entire document when you
are looking for specific information.
Copyright 2008, Analog Devices, Inc. All rights reserved. Analog Devices assumes no responsibility for customer product design or the use or application of
customers’ products or for any infringements of patents or rights of others which may result from Analog Devices assistance. All trademarks and logos are property
of their respective holders. Information furnished by Analog Devices applications and development tools engineers is believed to be accurate and reliable, however
no responsibility is assumed by Analog Devices regarding technical accuracy and topicality of the content provided in Analog Devices Engineer-to-Engineer Notes.
a
Table of Contents
Introduction .................................................................................................................................................................................................1
Table of Contents......................................................................................................................................................................................2
1 Brief Introduction to SDRAM .........................................................................................................................................................5
1.1 Basics of SDRAM............................................................................................................................................................................5
1.2 SDRAM Parameters in Blackfin Registers .....................................................................................................................6
1.2.1 EBCAW (SDRAM External Bank Column Address Width) ...................................................................................6
1.2.2 EBSZ (SDRAM External Bank Size)............................................................................................................................6
1.2.3 SDRAM Timing.........................................................................................................................................................................6
1.3 Multiprocessor Environment Options...............................................................................................................................7
1.3.1 BGSTAT (Bus Grant Status)..........................................................................................................................................7
1.3.2 PUPSD (Power-Up Startup Delay) ..............................................................................................................................7
1.3.3 CDDBG (Control Disable During Grant)................................................................................................................7
1.4 Mobile/ Low-Power SDRAM Options......................................................................................................................................8
1.4.1 PASR (Partial Array Self Refresh) .......................................................................................................................8
1.4.2 TCSR (Temperature-Compensated Self-Refresh) ...............................................................................................8
1.5 Options to Fit the SDRAM Timing......................................................................................................................................8
1.5.1 Blackfin Output / SDRAM Input Equation (Write) ........................................................................................9
1.5.2 Blackfin Input / SDRAM Output Equation (Read).........................................................................................10
2 SDRAM Initialization........................................................................................................................................................................10
2.1 SDRAM Initialization Via an Emulator and VisualDSP++ .XML Files........................................................10
2.2 Initialization Using Memory-Mapped Registers.....................................................................................................12
2.3 Initialization Using System Services........................................................................................................................15
2.4 SDRAM Initialization by the Values in the OTP Memory ..................................................................................17
2.5 Initializing Memory via Initialization Code Before Loading the Application..............................18
3 Using an .LDF File to Place Data and Program Code in Memory ...........................................................................21
4 SDRAM Hardware Design .....................................................................................................................................................................23
4.1 Connecting SDRAM to a Blackfin Processor (Schematics)................................................................................23
4.1.1 ADSP-BF53x Series Processors.................................................................................................................................23
4.1.2 ADSP-BF561 Processors (16-bit SDRAM)..............................................................................................................23
4.1.3 ADSP-BF561 Processors (32-bit SDRAM)..............................................................................................................24
4.2 High-Speed Design.....................................................................................................................................................................25
4.2.1 Effects that Impact Signal Quality...................................................................................................................25
4.2.2 Avoid Reflections...........................................................................................................................................................26
4.3 Design Guidelines for the SDRAM Connection..........................................................................................................27
4.3.1 Component Placement Considerations...................................................................................................................27
4.3.2 Using the Rounding Function of Your Layout Tool at Trace Edges ................................................28
4.3.3 Placing the VCC and GND Planes with as Little Distance as Possible.......................................28
Blackfin® Processor and SDRAM Technology (EE-326) Page 2 of 56

a
4.3.4 Insulating Critical Signals by Placing Them in the Inner Layers ..............................................29
4.3.5 Placing the Series Resistor Close to the SDRAM ......................................................................................29
4.3.6 Avoiding Trenches in the GND Plane...................................................................................................................29
4.3.7 Minimizing Back-Current Paths from Vias.......................................................................................................30
4.3.8 Make use of the drive strength control functionality ........................................................................31
4.3.9 Make use of the slew control functionality................................................................................................31
5 Using a Blackfin Processor with Less than 16 MB of SDRAM ..................................................................................31
5.1 System Settings..........................................................................................................................................................................32
5.2 Changing the .LDF File .........................................................................................................................................................32
5.2.1 Excursus: Background....................................................................................................................................................33
5.3 SDRAMs with 2 Banks ................................................................................................................................................................34
6 Increasing the SDRAM Performance of Your System........................................................................................................35
6.1 Optimal Multi-Bank Accesses .............................................................................................................................................35
6.2 Optimal Pages Accesses .........................................................................................................................................................36
Pages for the ADSP-BF53x Processors ..............................................................................................................................37
Pages for ADSP-BF561 Processors........................................................................................................................................38
6.3 SDRAM Performance Items for Core Accesses ............................................................................................................39
6.3.1 Code Overlays ....................................................................................................................................................................40
6.4 SDRAM Performance Items When Using Cache ..............................................................................................................40
Code........................................................................................................................................................................................................40
Data........................................................................................................................................................................................................40
7 Optimizing Power Consumption.....................................................................................................................................................41
7.1 Introduction: Power-Consumption Figures.................................................................................................................41
7.2 Tips for Lowering SDRAM Power Consumption ............................................................................................................41
7.3 Mobile SDRAM.................................................................................................................................................................................42
7.4 Going into Hibernate and Recover .................................................................................................................................42
Step 1 ...................................................................................................................................................................................................42
Step 2 ...................................................................................................................................................................................................43
Step 3 ...................................................................................................................................................................................................43
7.5 Structuring Data for Low Power Consumption..........................................................................................................43
Appendices ....................................................................................................................................................................................................44
Appendix A: Glossary .......................................................................................................................................................................44
Appendix B: Code Examples, Schematics, and Excursus..............................................................................................47
Initialization Code (Chapter 2)........................................................................................................................................47
Schematics to Interface SDRAM to the Blackfin Processor (Chapter 4)...................................................48
ADSP-BF561 in 16-Bit Mode ......................................................................................................................................................49
ADSP-BF561 in 32-Bit Mode ......................................................................................................................................................50
Excursus: Calculating Z0 (Chapter 4)..............................................................................................................................51

a
Calculating the Inductance of a Microstrip Trace................................................................................................52
Calculating the Capacitance .................................................................................................................................................52
IPC’s and Douglas Brooks’ approach of Z0 (Chapter 4)........................................................................................53
Microstrip Trace ...........................................................................................................................................................................53
Embedded Microstrip Trace ......................................................................................................................................................53
Stripline Trace..............................................................................................................................................................................53
Asymmetric Stripline Trace....................................................................................................................................................53
Dielectric Constants of Printed Circuit Boards (PCBs) (Chapter 4) .......................................................54
Several Commonly Available Woven Glass Reinforces Laminates......................................................................54
List of Non-Woven or Very-Low Glass Content Laminate Materials ..............................................................54
References ....................................................................................................................................................................................................55
Readings.........................................................................................................................................................................................................55
Document History ......................................................................................................................................................................................56

a
1 Brief Introduction to SDRAM
This section describes SDRAM (Synchronous Dynamic Random Access Memory) parameters and
SDRAM-related system setups. This provides a base on which to build a better understanding of the EBIU
(External Bus Interface Unit) register variables. Further, it will give you a foundation from which to base
other decisions. It is not a detailed discussion of SDRAM; refer to The ABCs of SDRAM (EE-126)[1] for
more details.
1.1 Basics of SDRAM

Although SRAM stores binary data using transistors, DRAM uses a capacitor and a transistor: The
capacitor is the storage itself; when charged, it will be a logical 1, otherwise a logical 0. The transistor
acts as a gate that controls access to the cell and traps the charge. The disadvantage of this technology is
that capacitors discharge over time.
To prevent the loss of data, the DRAM must be refreshed periodically to restore the charge on the memory
cells. Thus, a read and write operation must be performed to every memory location in the memory array
at least once during the refresh rate period, which is typically specified in milliseconds.
Figure 1. SDRAM
The DRAM cells are organized in an array of rows and columns. Every single cell can be accessed by a
well-defined row and column address. The row is often called the page, and the number of columns is
referred to as the page size. The address is time multiplexed; the row address is transmitted first, then the
column address is transmitted.
The /RAS (row access strobe) and by the /CAS (column access strobe) signals control the time
multiplexing of the row and column address. The /RAS signal indicates that a row address is available to
be loaded into the address buffer and decoded by the internal row address decoder. After a short delay, the
/CAS signal indicates that the column address, which is available in the address buffer, is forwarded to the
column address decoder.
If /WE is high, a read command is processed. The cells that are addressed by a row are read out
completely, amplified, and written back to the cells. The columns that are addressed are transmitted over
the data bus.

a
If the /WE is low, a write command is processed. The isolated writing of a single cell is not possible; thus,
a complete row is first read into a buffer. In the buffer, the elements to be changed are written, and the
complete row is written back to the array.
1.2 SDRAM Parameters in Blackfin Registers
1.2.1 EBCAW (SDRAM External Bank Column Address Width)

Often, the external bank column address width is called the page size of the SDRAM. Blackfin processors
support page sizes of 512 bytes (8 bits), 1 Kbytes (9 bits), 2 Kbytes (10 bits), and 4 Kbytes (11 bits).
Besides the byte address bit, which is the least significant bit (LSB), these bits are the least significant bits
of the logical address. Depended on the width of the column address, the row address and bank have
different bits positions in the logical address. For example, with a column addressing width (CAW) of
11 bits, the row address has its LSB at bit 12 of the logical 32-bit address[31:0]. This is a very important
parameter if you want to access the SDRAM consciously. The number of the row address depends on the
size of the SDRAM. To determine the column addressing width in the data sheet, find how many address
pins are dedicated to column addressing. For a detailed overview, refer to the EBIU chapter of the
Blackfin processor’s Hardware Reference.
1.2.2 EBSZ (SDRAM External Bank Size)
The SDRAM external bank size can be determined by the following formula:
Bank size x Number of Data pins x Number of Banks
MemorySize =
8
For example, the MT48LC32M16A2 has 8 Mbytes x 16 pins x 4 banks, which is 536,870,912 bits,
equaling 64 Mbytes.
To use less than 16 MB, refer to Using a Blackfin Processor with Less than 16 MB of SDRAM.
1.2.3 SDRAM Timing
Clock
Cmd ACTIV Pre ACTIV READ
Address Row Column
Data Data
t RAS t RP t RCD CL
Figure 2. SDRAM Timing

a
The SDRAM timing of SDRAM CAS Latency ( CL ), SDRAM bank activate command delay ( t RAS ),
SDRAM bank precharge delay ( t RP ), RAS to CAS delay ( t RCD ), and write to precharge delay ( t WR ) are
described in detail in the EBIU chapter of your processor’s Hardware Reference.
1.3 Multiprocessor Environment Options

Blackfin processors can be used in a multiprocessor environment. One approach to interfacing the
processors with one another is to share external memory and pass the messages via external memory to
the other processors. In order to provide arbitration functionality, Blackfin processors have dedicated pins
for this purpose.
To use the Blackfin processor in a multiprocessor environment with shared memory, you must connect the
/BR, /BG, and /BGH pins to provide an access control. When the external device requires access to the bus,
it asserts the Bus Request (/BR pin) signal. If no other request is pending, the external bus request will be
granted. The processor will three-state the data and address bus, and the bus grant (/BG pin) signal will be
asserted. When the bus is granted to another external device, any data or instruction fetch from the
external memory will stop the processor until the bus is released and the access can be executed. When
the external device releases /BR, the processor deasserts /BG and continues execution from the point at
which it stopped. The processor asserts the /BGH pin when it is ready to start another external port access,
but is held off because the bus was previously granted.
1.3.1 BGSTAT (Bus Grant Status)
When the bus has been granted, the BGSTAT bit in the SDSTAT register is set. This bit can be used by the
processor to check the bus status to avoid initiating a transaction that would be delayed by the external
bus grant.
1.3.2 PUPSD (Power-Up Startup Delay)
This option sets a delay of 15 system clock cycles for the power-up start sequence. If one processor passes
the SDRAM to another processor, it has to send the SDRAM into self-refresh mode. Once self refresh
mode is engaged, the SDRAM provides its own internal clocking, causing it to perform its own auto-
refresh cycles. The SDRAM must remain in self-refresh mode for a minimum period equal to t RAS .
The procedure for exiting self-refresh mode requires a sequence of commands. First, CLK must be stable
(stable clock is defined as a signal cycling within timing constraints specified for the clock pin) prior to
CKE going back high. When CKE is high, the SDRAM must have NOP commands issued for t XSR because
this length of time is required for the completion of any internal refresh in progress. To take the time
period t XSR into account, a delay of 15 cycles will occur until the Blackfin processor starts its power-up
sequence.
1.3.3 CDDBG (Control Disable During Grant)
If you are working in a multiprocessor environment with shared memory, where other devices than the
Blackfin processor have access to memory, this feature enables the external memory interface of the
processor to additionally three-state the address and data pins, its memory control pins (SRAS, SCAS, SWE,
SA10, and SCKE), and its clock pin (CLKOUT).

a
1.4 Mobile/ Low-Power SDRAM Options
Blackfin processors support mobile SDRAM chips (also called low-power SDRAM). Depending on the
series of Blackfin processors, you can use 3.3V, 2.5V, and 1.8V SDRAM. In order to use 2.5V and 1.8V
chips, verify in the processor’s data sheet (electrical specifications) that the processor is able to have an
output low voltage (maximum) as dictated by the SDRAM device data sheet. The mobile SDRAM
achieves low power consumption not only by its lower voltage, but also by its low currents. Another
advantage of mobile SDRAM is the ability to deactivate the self-refresh of the several banks you are not
using and to reduce the self-refresh rate by defining the temperature. These features enable you to
optimize power consumption for your specific application. If not otherwise stated in the Blackfin data
sheet, at 1.8V VDDEXT (the maximal frequency of the system clock) is limited to 100 MHz.
ADSP-BF51x ADSP-BF52x ADSP-BF533/2/1 ADSP-BF537/6/4 ADSP-BF539/8 ADSP-BF561
1.8V 9 9 9 8 8 8
2.5 V 9 9 9 9/8* 9/8* 9/8*
3.3 V 9 9 9 9 9 9
*Depending on selected SDRAM device.
Figure 3. Supported SDRAM voltages by processor series
To use the extended registers of mobile SDRAM, the EMREN bit must be set.
1.4.1 PASR (Partial Array Self Refresh)
Every refresh consumes power. If the application does not need to store the complete memory in special
modes, this feature provides a way to disable the refresh of several banks of the SDRAM. The benefit of
this feature is lower power consumption.
Take for example, an application that stores data (e.g., a picture) into SDRAM, does some signal
processing, transmits the data or stores it to a non-volatile memory(e.g., flash), and goes back into sleep
mode. Most of the SDRAM is needed only for the signal processing. The data is absolved after
processing, so the banks in which the data is placed does not need to be refreshed.
1.4.2 TCSR (Temperature-Compensated Self-Refresh)
In standard SDRAMs, the self-refresh rate is set to the worst-case scenario: A high temperature will cause
a higher discharge of the capacitors, which requires a higher refresh rate to maintain the data. But a higher
refresh rate will effect higher power consumption. By means of the TCSR bit, the user application is able to
set the self-refresh rate according to the temperature, which it is measuring.
1.5 Options to Fit the SDRAM Timing

Sometimes the timing parameter specification of an SDRAM does not fit the specification of the EBIU.
Therefore, the PLL control register provides an option to delay the output signal and shift the input latch
mechanism for 200 ps.

a
Figure 4. PLL_CTL register (copied from Hardware Reference)
Bit 6 and bit 7 (Figure 4) provide a delay to the SDRAM signals. Apply the output delay when the
specification of a write access will be violated. It will delay the data hold for 200 ps. The input delay must
be applied when the specification of a read access will be violated. It delays the latch of the incoming data
signal by 200 ps.
t sclk CLKOUT Period
t SSDAT Data Setup Before CLKOUT (BF data sheet)
t AC Access time from CLK (SDRAM data sheet)
t DH Input Hold time (SDRAM data sheet)
t HSDAT Data Hold after CLKOUT (BF data sheet)
t OH Output Hold time (SDRAM data sheet)
t OLZ Delay to the output high impedance from CLK (SDRAM data sheet)
t OHZ Delay from high impedance to signal from CLK (SDRAM data sheet)
1.5.1 Blackfin Output / SDRAM Input Equation (Write)

For SDRAM input, the following equation must be valid:
t DH < t HSDAT
t SCLK
t HSDAT
t DH
Figure 5. Timing

a
1.5.2 Blackfin Input / SDRAM Output Equation (Read)
This case applies especially when setting the SDRAM clock to 133 MHz, which places the timing
parameters at the limit of the specification.
t SCLK
Clock
DQ Data Data
tOLZ t SSDAT tOH tOHZ
t AC
Figure 6. Timing
To verify that the read operation processed correctly, ensure that t sclk > t SSDAT + t AC
2 SDRAM Initialization
SDRAM initialization impacts application performance and SDRAM power consumption. Therefore, a
better understanding of the settings is highly desirable. This section describes how to set up the EBIU
(External Bus Interface Unit) registers in order to run your application. In order to initialize the SDRAM,
use one of the following approaches:
SDRAM initialization via an emulator and VisualDSP++® .XML files
SDRAM initialization by setting the registers within the application
SDRAM initialization by using the VisualDSP++ system service model
SDRAM initialization by an initialization file before loading the actual application
SDRAM initialization by the values in the OTP (One Time Programmable) memory
2.1 SDRAM Initialization Via an Emulator and VisualDSP++ .XML Files

In the early stages of software development, you upload your software via an in-circuit emulator (ICE).
The emulator can set up the EBIU registers automatically when you upload a program to the processor.
To enable this functionality, from the VisualDSP++ Settings menu, chose Target Options. From the
resulting Target Options dialog box (Figure 7), select the Use XML reset values option.

a
Figure 7. Target Options dialog box
For VisualDSP++ release 4.5 or lower the values are taken from an .xml file (extensible markup
language) named ADSP-BF5XX-proc.xml (where XX stands for the architecture; e.g., ADSP-BF537-
proc.xml or ADSP-BF561-proc.xml). These files are located in the <install_path>\Analog
Devices\VisualDSP X.X\System\ArchDef directory. By changing these values, you can configure the
settings of the EBIU. Figure 8 shows a portion of a processor .XML file. For more information, refer to
“custom board support” in VisualDSP++ Help.
Figure 8: A portion of the ADSP-BF5xx-proc.xml file (VisualDSP++ 4.5)
The processor .xml file is read at startup only. Any editing while VisualDSP++ IDDE is up and running
will not take effect. Ensure that you have edited the values before invoking VisualDSP++ development
tools. When VisualDSP++ tools are running, the values in the .XML files are used to set the corresponding
SDRAM values of the Analog Devices EZ-KIT Lite® development board.
For VisualDSP++ release 5.0 or higher, it’s not recommended to modify the .XML
files in the ArchDef folder directly. Use the custom board support which is described
in the following paragraph.

a
With custom board support in VisualDSP++ release 5.0 or higher, it is easier to set reset values in an
application: This feature enables developers to have multiple sets of reset settings and to change between
them without having to start up VisualDSP++ tools again. Open a text editor and place code like the
following into a file and save it. In this example (Listing 1), the filename is
My_custom_board_reset_settings.xml:
<?xml version="1.0" standalone="yes"?>

<custom-visualdsp-proc-xml
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xsi:noNamespaceSchemaLocation="\Program Files\Analog Devices\VisualDSP
5.0\System\ArchDef\ADSP-custom-board.xsd"
processor-family="BLACKFIN"
file="My_custom_board_reset_settings.xml">



<custom-register-reset-definitions>
<register name="EBIU_SDRRC" reset-value="0x03A0" core="Common" />

<register name="EBIU_SDBCTL" reset-value="0x25" core="Common" />
<register name="EBIU_SDGCTL" reset-value="0x0091998D" core="Common" />
</custom-register-reset-definitions>
</custom-visualdsp-proc-xml>
Listing 1. Example custom board reset settings (VisualDSP++ 5.0)
To enable this feature in VisualDSP++ development tools, perform the following steps:
1. From the Settings menu, choose Session.
2. In the Session Settings dialog box, select Enable customizations.
3. In Custom board support file name, navigate to the file named
“My_custom_board_reset_settings.xml” that contains the SDRAM reset values.
Whenever the processor is reset by VisualDSP++ tools, these reset values will be set in the registers.
2.2 Initialization Using Memory-Mapped Registers

The classical approach to initializing the SDRAM controller is to assign the values directly to the
memory-mapped registers (MMRs).
As an example, we take a mobile SDRAM data sheet1.
1
Samsung K4M56163 R(B)N/G/L/F Mobile SDRAM

a
Figure 9. Example: portion of a data sheet for mobile SDRAM
Figure 10: Address configuration specification in the data sheet
Before setting the SDRAM registers, we must first configure the PLL. For our example, we are using a
system clock frequency of 133 MHz. First, we calculate the refresh rate:
f SCLK ⋅ t REF
RDIV = − (t RAS + t RP )
NRA
From the data sheet (-75), we get the following information:
fSCLK = 133 MHz
t REF = 64 ms
NRA = 8192
tRAS = 45 ns at 133 MHz: t RAS = 6 cycles
tRP = 18 ns at 133 MHz: t RP = 3 cycles
133 ⋅ 106 ⋅ 64 ⋅10−3
RDIV = − (6 + 3) = 1030.0625 ≈ 1030 in Hex : 0x406
8192
Further, we need to calculate the following values:
tRCD = 18 ns at 133 MHz: t RCD = 3 cycles
For Samsung specific devices, t WR is called t RDL .
t WR = t RDL = 2 cycles

a
Afterwards, we have to ensure that all derived parameters from the timings we specify are also within the
SDRAM specification. So we have to control:
t XSR = t RAS + t PR -> 6 cycles + 3 cycles = 9 cycles at 133 MHz : 67.5 ns
t RRD = t RCD + 1 -> 3 cycles + 1 cycle = 4 cycles at 133 MHz : 30 ns
t RC = t RAS + t PR -> 6 cycles + 3 cycles = 9 cycles at 133 MHz : 67.5 ns
t RFC = t RAS + t PR -> 6 cycles + 3 cycles = 9 cycles at 133 MHz : 67.5 ns
If we compare these values with the SDRAM’s data sheet, we notice we are within the specifications.
Additionally, we have to enable the extended register to set the temperature and the partial self-refresh.
The PASR can be set to following settings:
PASR_ALL – All four SDRAM banks refreshed in self-refresh
PASR_B0_B1 – SDRAM banks 0 and 1 are refreshed in self-refresh
PASR_B0 – Only SDRAM bank 0 is refreshed in self-refresh
Listing 2 shows the initialization in C.
//SDRAM Refresh Rate Setting
*pEBIU_SDRRC = 0x406;
//SDRAM Memory Bank Control Register
*pEBIU_SDBCTL = EBCAW_9 | //Page size 512
EBSZ_64 | //64 MB of SDRAM
EBE; //SDRAM enable
//SDRAM Memory Global Control Register
*pEBIU_SDGCTL = ~CDDBG & // Control disable during bus grant off
~FBBRW & // Fast back to back read to write off
~EBUFE & // External buffering enabled off
~SRFS & // Self-refresh setting off
~PSM & // Powerup sequence mode (PSM) first
~PUPSD & // Powerup start delay (PUPSD) off
TCSR | // Temperature compensated self-refresh at 85
EMREN | // Extended mode register enabled on
PSS | // Powerup sequence start enable (PSSE) on
TWR_2 | // Write to precharge delay TWR = 2 (14-15 ns)
TRCD_3 | // RAS to CAS delay TRCD =3 (15-20ns)
TRP_3 | // Bank precharge delay TRP = 2 (15-20ns)
TRAS_6 | // Bank activate command delay TRAS = 4
PASR_B0 | // Partial array self refresh Only SDRAM Bank0
CL_3 | // CAS latency
SCTLE ; // SDRAM clock enable
Listing 2. Initialization via registers
The code in assembly language can be found in Initialization Code (Chapter 2).
What will happen if we need to change the PLL settings later in the application? To obtain the best
memory performance, we have to change our settings as well. If memory performance does not matter, we
have to calculate the values at the worst-case scenario.

a
Examples of initializing the SDRAM on the ADSP-BF537 EZ-KIT Lite (in assembler and C) and on the
ADSP-BF561 EZ-KIT Lite (in C) are included together with this EE-Note.
2.3 Initialization Using System Services

Another way to initialize the SDRAM is to handle the setup in the application itself. The most
comfortable way to change the EBIU settings is by means of system services. The main advantage of
using the EBIU system service is that you can set the EBIU without any calculations. And second benefit
is that the changes to the EBIU are in accordance to the PLL changes, so you do not need to be concerned
about the actual system clock frequency.
The initialization is done by the adi_ebiu_init function. We will take the same SDRAM as in the
section above.
To perform an initialization using the system services, we have to pass the SDRAM parameters to the
system. Therefore, we are using a command structure called ADI_EBIU_COMMAND_PAIR. (Table 1).
To initialize the structure in our example, we set the variables (for the -75) as shown in Listing 3.
// set the sdram to 64 MB
ADI_EBIU_SDRAM_BANK_VALUE bank_size = {0, ADI_EBIU_SDRAM_BANK_64MB};
//set the sdram to 9 bit Column Address Width (like we see in the Address
//Configuration)
// A0-A8 -> 9bits Column Address
ADI_EBIU_SDRAM_BANK_VALUE bank_caw = {0,
(ADI_EBIU_SDRAM_BANK_SIZE)ADI_EBIU_SDRAM_BANK_COL_9BIT}; // 9bit CAW
//set the twr to 2 sclk + no time

ADI_EBIU_TIMING_VALUE MyTWR = { 2, // 2 cycle
{0, ADI_EBIU_TIMING_UNIT_PICOSEC}}; // 0 ns
//set refresh rate to 64ms at 8192 rows as seen in the data sheet
ADI_EBIU_TIMING_VALUE Refresh = { // SDRAM Refresh rate:
8192, // 8192 cycles
{64, ADI_EBIU_TIMING_UNIT_MILLISEC }}; // 64ms
//set TRAS to 45ns

ADI_EBIU_TIME MyTRAS = {45, ADI_EBIU_TIMING_UNIT_NANOSEC};
//set TRP to 18ns
ADI_EBIU_TIME MyTRP = {18, ADI_EBIU_TIMING_UNIT_NANOSEC};
//set TRCD to 18ns
ADI_EBIU_TIME MyTRCD = {18, ADI_EBIU_TIMING_UNIT_NANOSEC};
//set CAS threshold frequency
u32 MyCAS = 133;
//Enable Extended Mode Register because we are using Mobile SDRAM
ADI_EBIU_SDRAM_EMREN MyEMREN = ADI_EBIU_SDRAM_EMREN_ENABLE;
//Refresh only the first bank
ADI_EBIU_PASR MyPASR = ADI_EBIU_PASR_INT0_ONLY;
//Temperature Compensation at 85°C

ADI_EBIU_SDRAM_TCSR MyTCSR = ADI_EBIU_SDRAM_TCSR_85DEG;
//we don't have any registered buffer
ADI_EBIU_SDRAM_EBUFE MyEBUFE = ADI_EBIU_SDRAM_EBUFE_DISABLE;
//no fast-back-to-back read/write
ADI_EBIU_SDRAM_FBBRW MyFBBRW = ADI_EBIU_SDRAM_FBBRW_DISABLE;

a
//Do not disable the control during bus grant
ADI_EBIU_SDRAM_CDDBG MyCDDBG = ADI_EBIU_SDRAM_CDDBG_DISABLE;
//We don't need any delay at Power Up
ADI_EBIU_SDRAM_PUPSD MyPUPSD = ADI_EBIU_SDRAM_PUPSD_NODELAY;
//Do first the refresh
ADI_EBIU_SDRAM_PSM MyPSM = ADI_EBIU_SDRAM_PSM_REFRESH_FIRST;
Listing 3.Initialization of the EBIU structure
Table 1. Overview of EBIU commands
After setting up the parameters, we need to initialize the service. Therefore, we bundle the parameters to a
command table and afterwards call the adi_EBIU_init function.
ADI_EBIU_COMMAND_PAIR Sdram_Values[] = {
{ ADI_EBIU_CMD_SET_SDRAM_BANK_SIZE, (void*)&bank_size },
{ ADI_EBIU_CMD_SET_SDRAM_BANK_COL_WIDTH, (void*)&bank_caw },
{ ADI_EBIU_CMD_SET_SDRAM_CL_THRESHOLD, (void*)&MyCAS },
{ ADI_EBIU_CMD_SET_SDRAM_TRASMIN, (void*)&MyTRAS },
{ ADI_EBIU_CMD_SET_SDRAM_TRPMIN, (void*)&MyTRP },
{ ADI_EBIU_CMD_SET_SDRAM_TRCDMIN, (void*)&MyTRCD },
{ ADI_EBIU_CMD_SET_SDRAM_TWRMIN, (void*)&MyTWR },
{ ADI_EBIU_CMD_SET_SDRAM_REFRESH, (void*)&Refresh },
{ ADI_EBIU_CMD_SET_SDRAM_FBBRW, (void*)&MyFBBRW },
{ ADI_EBIU_CMD_SET_SDRAM_EMREN, (void*)&MyEMREN },
{ ADI_EBIU_CMD_SET_SDRAM_PASR, (void*)&MyPASR },
{ ADI_EBIU_CMD_SET_SDRAM_TCSR, (void*)&MyTCSR },
{ ADI_EBIU_CMD_SET_SDRAM_EBUFE, (void*)&MyEBUFE },
{ ADI_EBIU_CMD_SET_SDRAM_CDDBG, (void*)&MyCDDBG },
{ ADI_EBIU_CMD_SET_SDRAM_PUPSD, (void*)&MyPUPSD },
{ ADI_EBIU_CMD_SET_SDRAM_PSM, (void*)&MyPSM },
{ ADI_EBIU_CMD_END, 0}
};
//Init the service and ensure that the Refresh rate is reset if fsclk is changing
Result = adi_ebiu_Init( Sdram_Values, true); // true enables automatic adjustment
Listing 4. Initialization of the EBIU

a
Code examples that demonstrate the use of system services for the ADSP-BF537 and the ADSP-BF561
processors are located in the .ZIP file attached to this EE-Note.
2.4 SDRAM Initialization by the Values in the OTP Memory

Devices with One-Time-Programmable (OTP) memory (ADSP-BF51x, ADSP-BF52x and ADSP-BF54x
processors) offer an additional approach to initializing the EBIU register settings. During booting, the
boot ROM program is able to initialize the EBIU settings by the programmed values in the OTP. The OTP
consists of a region called the Preboot Settings (PBS) block, which must be programmed via the
OTP_write function.
The PBS02L page contains the EBIU settings. Refer to your processor’s Hardware Reference for an
overview.
After setting up the EBIU control registers, the boot process accesses the SDRAM at address 0x00000000
in order to initialize the SDRAM. By default, setting a read access will be performed. Since a read access
consumes more time than a write access, this access can be customized by the
OTP_EBIU_POWERON_DUMMY_WRITE bit (Figure 11), which replaces the read access with a write access and
saves time. By using this option, you must ensure that no important data is stored at this address before
going to reset or hibernate.
Figure 11. Dummy write option
/* Declare local variable */

DU64 Data;
.
.
/* Enable the EBIU Init settings */
Data.h = 0x04000000 | //use the content of PBS02L for the EBIU
... | //other settings
Data.l = ...
otp_write(0x18, OTP_LOWER_HALF, &Data);
/* Assign SDRAM values */
/* The low data byte is the EBIU_SDGCTL register */

Data.l = ~CDDBG & // Control disable during bus grant off

a
SCTLE;
/* The upper 32 bit are EBIU_SDBCTL(0-15) and EBIU_SDRRC (16-27). Further we will
Set the dummy write bit (32), which will speed up the initialization */
Data.h = 0x80000000 | // dummy write bit

(0x406)<<16 | // Refresh rate
EBCAW_9 | //Page size 512
EBE; //SDRAM enable
;
otp_write(0x1A, OTP_LOWER_HALF, &Data);
Listing 5. Initialization via PBS
2.5 Initializing Memory via Initialization Code Before Loading the Application
If you want to place instruction and data sections into your SDRAM at initialization time, you must use an
initialization file that initializes the SDRAM before the application is loaded. Therefore, start a project
and code an initialization file. Build this project into a .DXE file. The initialization .DXE file can be
included into a loader file of your actual application via the Project Options dialog box
(Project:Load:Options page).
Figure 12. Specifying project load options

a
Listing 6 shows an example of how initialization code can be implemented.
#include <defBF537.h>
.section program;
//save all registers on the stack
[--SP] = ASTAT;
...
[--SP] = L3;
//Setup the PLL Settings

//Ensure no other interrupt will disturb us
CLI R1;
//Setup the voltage regulator
P0.L = lo(VR_CTL);
P0.H = hi(VR_CTL);
R0 = 0x40DB (z);
w[P0] = R0;
//Wait the VR settings are done

idle;
//Setup the DIV of ssclk and cclk
P0.L = lo(PLL_DIV);
R0 = 0x0003 (z);
w[P0] = R0;
//Setup wait cycles until PLL is set

P0.L = lo(PLL_LOCKCNT);
R0 = 0x200 (z);
w[P0] = R0;
//Setup the PLL

P0.L = lo(PLL_CTL);
R0 = 0x2000 (z);
w[P0] = R0;
//Wait the PLL settings are done

idle;
//restore interrupts
STI R1;
//Our program needs 4 MBs of RAM

P0.L = lo(EBIU_AMGCTL);
P0.H = hi(EBIU_AMGCTL); //Asynchronous Memory Global Control Register
//Uncomment your setting
R0 = 0x00F0
// |AMBEN_NONE(Z); //No Asynchronous Memory
// |AMBEN_B0(Z); //1MB Asynchronous Memory
// |AMBEN_B0_B1(Z); //2MB Asynchronous Memory
// |AMBEN_B0_B1_B2(Z); //3MB Asynchronous Memory
|AMBEN_ALL(Z); //4MB Asynchronous Memory
W[P0] = R0;
//Setup the SDRAM

P0.H = hi(EBIU_SDRRC);
P0.L = lo(EBIU_SDRRC);
R0 = 0x406 (z);
w[P0] = R0;

a
P0.H = hi(EBIU_SDBCTL);
P0.L = lo(EBIU_SDBCTL);
R0 = EBCAW_9 | //Page size 512
EBE; //SDRAM enable
w[P0] = R0;

P0.H = hi(EBIU_SDGCTL);
P0.L = lo(EBIU_SDGCTL);
R0.H= hi( ~CDDBG & // Control disable during bus grant off
SCTLE ); // SDRAM clock enable
R0.L= lo( ~CDDBG & // Control disable during bus grant off
SCTLE ) ; // SDRAM clock enable
[P0] = R0;
ssync;
//Restore registers from the stack
L3 = [SP++];
...
ASTAT = [SP++];
RTS;
Listing 6. Initialization via an initialization file

a
3 Using an .LDF File to Place Data and Program Code in Memory
If the SDRAM is initialized before the application is loaded (see SDRAM Initialization by the Values in
the OTP Memory), you can place data and code into the memory sections. Therefore, you must define a
memory region in the Linker Description File (.LDF) and inform the linker as to the data or code that is to
be placed into memory.
Let’s assume we want to place data in SDRAM bank 1. If we take a closer look into a standard Linker
Description File (.LDF) which is generated by the VisualDSP++ Expert Linker wizard (in this example we
will have 512 MB of SDRAM), we see the internal SDRAM banks are named separately:
MEMORY
{ MEM_SYS_MMRS { TYPE(RAM) START(0xFFC00000) END(0xFFDFFFFF) WIDTH(8) }
[…]
MEM_SDRAM0_BANK0 { TYPE(RAM) START(0x00000004) END(0x07ffffff) WIDTH(8) }
MEM_SDRAM0_BANK1 { TYPE(RAM) START(0x08000000) END(0x0fffffff) WIDTH(8) }
MEM_SDRAM0_BANK3 { TYPE(RAM) START(0x18000000) END(0x1fffffff) WIDTH(8) }
} /* MEMORY */
Listing 7.Assigning memory
We will find the memory label “MEM_SDRAM0_BANK1” in the “SECTIONS” area again:
PROCESSOR p0
{
OUTPUT($COMMAND_LINE_OUTPUT_FILE)
RESOLVE(start, 0xFFA00000)
KEEP(start, _main)
SECTIONS
{ [...]
sdram0_bank1
{
INPUT_SECTION_ALIGN(4)
INPUT_SECTIONS($OBJECTS(sdram0) $LIBRARIES(sdram0))
INPUT_SECTIONS($OBJECTS(sdram0_bank1) $LIBRARIES(sdram0_bank1))
INPUT_SECTIONS($OBJECTS(sdram0_data) $LIBRARIES(sdram0_data))
INPUT_SECTIONS($OBJECTS(cplb) $LIBRARIES(cplb))
INPUT_SECTIONS($OBJECTS(data1) $LIBRARIES(data1))
INPUT_SECTIONS($OBJECTS(voldata) $LIBRARIES(voldata))
INPUT_SECTIONS($OBJECTS(constdata) $LIBRARIES(constdata))
INPUT_SECTIONS($OBJECTS(cplb_data) $LIBRARIES(cplb_data))
INPUT_SECTIONS($OBJECTS(.edt) $LIBRARIES(.edt))
INPUT_SECTIONS($OBJECTS(.cht) $LIBRARIES(.cht))
} > MEM_SDRAM0_BANK1
[...]
} /* SECTIONS */
} /* p0 */
Listing 8. Specifying memory sections

a
We see that several libraries and objects are mapped into this memory space. You can map an object or
library to more than one memory section. In this case, the linker decides which part of the data or code is
placed into each section of memory.
To place data explicitly in one memory section if using C/C++, use the pragma section directive. In our
case, we want to place the global variable x into internal SDRAM bank 1. The label sdram0_bank1 is
only mapped to internal SDRAM bank 1 and no other section. So we will use this label to map our
variable.
//define a variable which lies in SDRAM Bank 1
#pragma section ("sdram0_bank1")
long int x = 0;
We can do the same with our instruction code by assigning the prototypes of our functions to a section:
//define a function prototype of a function which lies in SDRAM Bank 1
#pragma section ("sdram0_bank1")
void foo();
As described in Increasing the SDRAM Performance of Your System, sometimes it is useful to map a data
section manually to prevent delays caused by opening and closing a page. Therefore, we define our own
memory section in the .LDF file:
//define my own page
MEM_SDRAM0_BANK3_PAGE0 { TYPE(RAM) START(0x03000000) END(0x030007FF) WIDTH(8) }
In the SECTIONS part of the Linker Description File, we will link all objects labeled MyDefinedMemory
into this memory space.
sdram0_bank3_page_0
{
INPUT_SECTIONS($OBJECTS(MyDefinedMemory) )
} > MEM_SDRAM0_BANK3_PAGE0
Afterwards, in our C/C++ file, we place our array into this memory section.
//define an array which lies in my own defined memory space (Bank 3 Page 0)
#pragma section ("MyDefinedMemory")
long int MyArray[100];

a
4 SDRAM Hardware Design
Since the SDRAM is driven by frequencies greater than 50 MHz, the hardware layout must fulfill the
requirements of high-speed design. Nowadays, hardware designs satisfy many standards concerning EMI
and EMC. They have to ensure the signal integrity at high frequencies, and many of them are set in a low
power environment. Therefore, the right printed circuit board (PCB) design is a key factor. This section
explains how to design the connection between the Blackfin processor and the SDRAM on a PCB.
4.1 Connecting SDRAM to a Blackfin Processor (Schematics)

Blackfin processors provide a glueless interface to SDRAM. Depending on the Blackfin processor,
SDRAMs with a power supply requirement of 1.8V to 3.3V are supported.
A common design mistake occurs when is the Blackfin processor address pins are not connected to the
SDRAM correctly.
The address lines must be connected as described next.
4.1.1 ADSP-BF53x Series Processors
Connect Blackfin processor’s ADDR1 to the SDRAM A0, ADDR2 to A1, etc.
Do not use the ADDR11 of the Blackfin; connect SA10 to A10
Connect ADDR18 to the SDRAM’s BA0
Connect /ABE0 to the DQML pin (for 16-bit SDRAM) or to DQM of the chip(s) connected to D0-D7
Connect /ABE1 to the DQMH pin (for 16-bit SDRAM) or to DQM of the chip(s) connected to D8-D15
Figure 13. ADSP-BF53x processors: connections between the Blackfin processor and SDRAM
4.1.2 ADSP-BF561 Processors (16-bit SDRAM)

Connect the Blackfin processor’s ADDR2 to the SDRAM’s A1, ADDR3 to A2, etc.
Connect the Blackfin processor’s SDQM3 to the SDRAM’s A0
Do not use the ADDR11 of the Blackfin processor; connect SA10 to A10

a
Connect SDQM0 to DQML or to DQM (see above)
Connect SDQM1 to DQMH or to DQM (see above)
Connect the /SMSx to the /CS lines
(when using x16 and are using more than one SDRAM: to each chip one /SMS line)
(when using x8 and are using more than two SDRAMs: to each pair of two one /SMS line)
BF561 SDRAM
SDQM3 A0
A2-A10, A12-A17 A1-A9, A11-A16
SA10 A10
A18 A19 BA0 BA1
D0-D15 DQ0-DQ15
SDQM0 DQML
SDQM1 DQMH
SRAS RAS
SCAS CAS
SWE WE
SMS0-SMS3 CS
SCKE CKE
CLK0_OUT SCLK CLK
Figure 14. ADSP-BF561 processors: connections between the Blackfin processor and 16-bit SDRAM
4.1.3 ADSP-BF561 Processors (32-bit SDRAM)

Connect the Blackfin processor’s ADDR2 to the SDRAM’s A0, ADDR3 to A1, etc.
Do not use the ADDR12 of the Blackfin; connect SA10 to A10
Connect SDQM0 to DQML of SDRAM1(D0-D15)
Connect SDQM1 to DQMH of SDRAM1
Connect SDQM3 to DQML of SDRAM2(D16-D31)
Connect SDQM4 to DQMH of SDRAM2
If you are using 8-bit SDRAM
Connect SDQM0 to DQM of SDRAM1(D0-D7)
Connect SDQM0 to DQM of SDRAM2(D8-D15)
Connect SDQM0 to DQM of SDRAM3 (D16-D23)
Connect SDQM0 to DQM of SDRAM4 (D24-D31)

a
Figure 15. ADSP-BF561 processors: connections between the Blackfin processor and 32-bit SDRAM
Another problem is caused when the /BR pin is left floating. The Blackfin processor interprets the signal
as bus request and will answer with the /BG signal, which will block the parallel bus for an infinite time.
So always put a pull-up resistor on the /BR pin.
Use decoupling capacitors to decouple the SDRAM power supply.
Add a series resistor close to each data pin of the SDRAM. The next section explains how to determine
the resistance.
4.2 High-Speed Design

The layout of the SDRAM connection is a critical factor, especially on low-power designs. This section
explains how to optimize the design of the SDRAM layout to fit your application’s requirements. The
most critical connections are the clock, the lower address lines, the DQM, and the data lines.
4.2.1 Effects that Impact Signal Quality
This section describes effects that are influenced by your hardware design.
Reflection
If the impedance of a connection line is equivalent to the input resistance of the receiver, the energy will
be fully absorbed by the resistance. Otherwise, the transmission’s energy will be thrown back. This will
interfere with the desired signal by superposition.
Coupling
The currents conducted through different traces influence each other. When a changing current flows
down trace A, it creates a changing magnetic field that couples into trace B. The coupling generates a
current in trace B that is dependent upon the coupling factor. The inducted current’s direction is opposite
to the current in trace A; this effect is negative if two signal traces influence each other (crosstalk). But
the effect can also be positive when the influence is between the signal line and its return line (GND). The
coupled signal helps to boost the return signal, and the returning signal boosts the primary signal.

a
4.2.2 Avoid Reflections
Calculate the characteristic impedance to add the correct termination resistor.
There are several ways to terminate a transmission line:
Series termination
Parallel termination
Thevenin termination
Termination by diodes
AC termination
The most important techniques for SDRAM devices are described next.
Series Termination
For SDRAM, use a series termination (Figure 16). Place the series resistor close to the output pin of the
transmitter. The advantage is that there will not be any DC current draw (like if you are using a parallel
termination). This is essential for low-power designs. The disadvantage is that there will be a nearly 100%
refection at the receiver, which is thrown back to the transmitter. Since the lengths of SDRAM traces are
short and the traces of one signal from the Blackfin processor to each of the SDRAM chips have nearly
the same length, this effect will not impact the SDRAM’s functionality.
Figure 16. Using series termination
To avoid a second reflection from the driver (transmitter), the resistor must have the right value. The
value for the series termination resistor has to be set so that the sum of it and the output impedance of the
driver equals the impedance of the trace. As an equation, we get the following:
R S = Z 0 − Z OUT , whereby ZOUT is the output impedance of the transmitter
The read command is more critical than a write command. Thus, place the resistor of the data line as close
as possible to the SDRAM data pin.
Parallel Termination
The alternative is to use a parallel termination (Figure 17). As mentioned earlier, this is not necessary for
standard SDRAM when you follow the design guidelines at the end of this chapter.

a
Figure 17. Using parallel termination
ZL − Z0
Reflection coefficent ρ =
ZL + Z0
To keep the reflection as low as possible, the parallel termination resistor (ZL) should be equivalent to Z0.
Consider whether you even need termination. An additional resistor emits further EMI.
Note that for ADSP-BF51x and ADSP-BF52x processors termination is mandatory.
Consult the datasheet for more details.
4.3 Design Guidelines for the SDRAM Connection

With regard to EMC and signal integrity, the following design guidelines are recommended. When you
start your SDRAM PCB layout, do not treat all signals the same, consider the importance of each signal
and place the traces of the most critical signals first. The following succession is suggested:
1. Clock distribution
2. Data lines and DQM lines, command lines including SA10
3. Address lines
4. Other signals (e.g., CKE)
4.3.1 Component Placement Considerations
Consider the following points while laying out your PCB:
Place the SDRAM chips close to the Blackfin processor.
Keep the traces as short as possible.
When you are distributing a signal, all traces should have the same path length (Figure 18) to the
devices (if possible). Avoid loops like the one shown in Figure 19.
Figure 18. Right: spread traces and make their lengths equal

a
S
D
R
A
M
S
D
R
A
M
Figure 19. Wrong: traces that loop around
4.3.2 Using the Rounding Function of Your Layout Tool at Trace Edges
Figure 20 shows a PCB trace edge that does not use rounding. Figure 21 shows the same trace edge when
the rounding feature is enabled.
Figure 20. Wrong: Trace edges lack rounding
Figure 21. Right: Trace edges are rounded
4.3.3 Placing the VCC and GND Planes with as Little Distance as Possible
Figure 22 shows a 4-layer PCB, which does not insulate the critical signals. Figure 23 shows proper
insulation of a 4-layer PCB..
critical signals
other signals
VCC
GND
Figure 22. Wrong: VCC and GND planes are too far apart
other signals
VCC
GND
critical signals
Bold layer
As thin as possible
Figure 23. Right: VCC and GND planes are close together

a
4.3.4 Insulating Critical Signals by Placing Them in the Inner Layers
Figure 24 shows a 6-layer PCB, in which critical signals are not insulated. Figure 25 shows proper
insulation.
Figure 24. Wrong: critical signals not properly insulated
Figure 25. Right: minimize the distance between critical signals and the ground plane
4.3.5 Placing the Series Resistor Close to the SDRAM

Figure 26 shows a signal path that with too many vias; the series resistor is too far from the SDRAM.
Figure 27 shows a short signal path without vias and a series resistor that is next to the SDRAM.
Figure 26. Wrong: too many vias in critical signal path and series resistor is too far away
Figure 27. Right: the series resistor is close to the SDRAM
4.3.6 Avoiding Trenches in the GND Plane

Figure 28 shows a GND plane with a trench. The GND plane in Figure 29 avoids having a trench.

a
Figure 28. Wrong: ground plane with a trench
Figure 29. Right: ground plane without a trench
4.3.7 Minimizing Back-Current Paths from Vias

If you are not able to avoid a direction change in a via from one layer to another, try to minimize the way
of the back current. Figure 30 shows a signal path that runs in two directions from the via. The direction
change in Figure 31 is better.
Figure 30. Wrong: layer-to-layer direction change
Figure 31. Right: avoiding a direction change

a
4.3.8 Make use of the drive strength control functionality
To reduce electromagnetic emissions, lower the drive strength of the EBIU pins. But keep an eye on the
signals: We have to ensure that we supply enough current to load the capacity of lines and pins within the
timing specifications. 00b within the bit fields indicates low drive strength (see data sheet), 01b indicates
high drive strength.
Figure 32. The Drive Strength control register of the BF52x
4.3.9 Make use of the slew control functionality

A smoother transition also helps reducing electromagnetic emissions. Verify again whether you stay
within the timing specifications after changing it to slower slew rate. 00b indicates a faster slew rate, 01b
a slower slew rate.
Figure 33. The Slew Rate register of the BF52x
5 Using a Blackfin Processor with Less than 16 MB of SDRAM

Using less than 16 MBs (128 Mbits) of SDRAM is especially important for low-power applications. This
section provides guidance for applications that use less than 16 MBs.

a
5.1 System Settings
On the hardware side, there are no special settings. Just connect the address lines as described in SDRAM
Hardware Design.
The first step in using less than 16 MB is setting the SDRAM external bank size bits of the EBIU_SDBCTL
(SDRAM memory bank control) register to 16 Mbytes. This configures the Blackfin processor's internal
address to expect 16 Mbytes, and the address space will be fragmented as shown in Figure 34.
Figure 34. Address space
The “aliased” address space’s content is a copy of the according bank, and every write access to this space
results in a write access into the “real” bank. This must be considered by the Linker Description File
(.LDF), or the Blackfin processor will place instructions and data into addresses that do not exist.
This type of address failure cannot be detected by the Blackfin processor. There is no
functionality that tests whether an address is valid. This will cause failures later in your
application when the core is tries to read from a non-existent address space, getting dummy
values, and then interprets this as valid instruction code or data.
5.2 Changing the .LDF File

First, ensure that the .LDF file will not be changed by the Expert Linker wizard. Therefore, open the
Project Options dialog box (Project -> Project Options) to the Remove Startup Code/LDF page
and select the Leave the files in the project, but stop regenerating them option (Figure 35).

a
Figure 35. Ensuring that the .LDF file will not be changed
Doing so allows you to change the .LDF file manually.

For example, if we use a 64-Mbit SDRAM (8 MB), the address space in the Linker Description File has:
MEM_SDRAM0_BANK0 { TYPE(RAM) START(0x00000000) END(0x001FFFFF) WIDTH(8) }
MEM_SDRAM0_BANK3 { TYPE(RAM) START(0x00C00000) END(0x00DFFFFF) WIDTH(8) }
Figure 36. Ensuring that the .LDF file will not be changed
As shown in Figure 36, there are gaps of 2 MB in our address space.

5.2.1 Excursus: Background
The .LDF file specifies where code and data are placed in memory space. The EBIU settings configure the
SDRAM controller of the Blackfin processor and specify size, timing, and features of the SDRAM. Since
the EBIU settings cannot set to that of an 8 MB SDRAM, we have to set the SDRAM size to 16 MB.
What does this mean to the addressing? We are using an SDRAM with a column address width (CAW) of
10 bits, which means we can address 2^10 = 1024 columns. With each column and row address, we are
addressing 2 bytes (for x16 SDRAM). Now, we have set the RAM to 16 MBs (=16777216 bytes), which
means we have a row address width of 13 (memory size / (data width * 2^addresses) = 16777216/(2 bytes
* 2^10)= 8192= 2^13), but we are using only a 8 MB RAM, which has a row address width of 12. But the
controller has calculated a row address width of 13. The row address and the column address are sent to
the SDRAM time multiplexed. Looking at your SDRAM, you see 12 address lines and 2 bank address
lines. But the controller has calculated 13 and is using 13. Since the 13th address line is not connected, you
will address the same physical address independent of the state of bit 13 of the row address. This is why
the memory space is mirrored.
For example, row address 0x1000 will access the same data as row address 0x0000. It is critical when you
place something into aliased memory space, because you overwrite something in the other address space.

a
Therefore, we define the .LDF file that way, so we will not place anything into the mirrored RAM:
MEM_SDRAM0_BANK3 { TYPE(RAM) START(0x00C00000) END(0x00DFFFFF) WIDTH(8) }
5.3 SDRAMs with 2 Banks

Several SDRAMs have only two banks. The hardware connection has to be like Figure 37.
Addr[1] A0
Blackfin SDRAM
Addr[18] BA
Addr[19]
Figure 37. Hardware connection for SDRAM with two banks
BA is the bank selection pin of the SDRAM. It must be connected with the Addr[18] of the Blackfin
processor. Leave Addr[19] floating. Connect the other addresses as described in SDRAM Hardware
Design. Set the EBIU_SDBCTL (SDRAM memory bank control) register to 16 Mbytes. The logical address
space will be fragmented as shown in Figure 38.
Figure 38. Fragmented logical address space
So we have to use adjust the .LDF file again to set up the memory space. Proceed as described above.
Figure 39 shows an example memory space for a 16-Mbit (2 MB) SDRAM.
Figure 39. Example: Memory space for a 16-Mbit (2-MB) SDRAM
As shown, there is a 3-MB gap in our address space.

a
6 Increasing the SDRAM Performance of Your System
In many applications, execution time is one of the key factors. The placement of data and instructions can
have a significant impact to the processing speed of your application. This section presents an overview of
the possible ways of increasing SDRAM performance. For a system approach, refer to System
Optimization Techniques for Blackfin Processors (EE-324)[2].
6.1 Optimal Multi-Bank Accesses

It is time-consuming to open a page via the activate command or close a page via the precharge command.
Thus, reducing page changeovers results in better SDRAM performance.
A case that applies to many applications is coping data from one array to another via memory DMA. The
problem occurs when both arrays are on the same internal bank. If this happens within a page, it will not
be a problem, but if the dimension of the two arrays exceeds the page size, the DMA will access at least
two pages.
Figure 40 shows how the DMA works within a bank (single bank access).
Figure 40. Accesses to one memory bank
The data is placed on different pages on the same bank. An activate and precharge must be executed after
each switch between source and destination DMA. Additionally, we need to take into account that the
core (or the cache) is may access the bank as well. There is a delay every time between the reads and the
writes. This delay can be enlarged by the internal DMA architecture: The DMA is designed as a feedback
control state machine, which introduces additional wait states under special circumstances.
To avoid such a time-consuming case, organize the memory in a way that allows inter-bank DMA copies.
Figure 41 shows such an approach.

a
Figure 41. Multi-banking approach via DMA and core
The core gets its code from bank 0, and the MDMA transfer runs from bank 3 to bank 2. Figure 42 shows
the sequence of the data transfer between two banks. As shown, the number of precharge commands and
activate commands decreases significantly. As discussed earlier, because precharge and activate
commands are time-intensive procedures, this technique saves a lot of time.
Figure 42. Intelligent bank accesses
6.2 Optimal Pages Accesses

This section describes how to improve page accesses. If you are able to keep accesses within a page, now
further activate and precharge commands are necessary (besides those that are issued to do the refresh).
The following description shows how to access pages separately.
Open the Remove Startup Code/LDF page of the Project Options dialog box (Project -> Project
Options) and select the Leave the files in the project, but stop regenerating them option.
Also, select the Remove the generated LDF check box. Be sure to confirm the selections by clicking the
OK button.

a
Figure 43. Project Options dialog box settings
Now we are able to change the Linker Description File (.LDF).

Pages for the ADSP-BF53x Processors
ADSP-BF53x processors have a 16-bit memory interface. The address mapping scheme is defined in
Figure 44.
Figure 44. ADSP-BF53x memory mapping scheme
There are different page sizes, depending on the SDRAM being used. Table 2 shows 16-bit EBIU page
size with regard to the EBCAW bits.
EBCAW Page bytes in Hex
8 bits 0x200
9 bits 0x400
10 bits 0x800
11 bits 0x1000
Table 2. Page sizes for 16-bit EBIU
Consider a column address width of 10 bits. Every page has 0x800 bytes, and we are able to define our
memory mapping in the .LDF file (Listing 9).

a
MEMORY
{...
/* We define 10 pages in bank 0 of a 10-bit memory of a BF53x
SDRAM_BANK_0_PAGE_0 { TYPE(RAM) START(0x00000000) END(0x000007FF) WIDTH(8) }
SDRAM_BANK_0_PAGE_1 { TYPE(RAM) START(0x00000800) END(0x00000FFF) WIDTH(8) }
//if we would define a section for each page we have to define 8192...
//so we define sections only for the amount of pages which are performance
//relevant
SDRAM_BANK_0_OTHER{ TYPE(RAM) START(0x00005000) END(0x00FFFFFF) WIDTH(8) }
/* the pages on the second bank… */
...
}
PROCESSOR p0
{
SECTIONS
{ ...
sdram0_page_0
{
INPUT_SECTIONS($OBJECTS(sdram0page0) )
} > SDRAM_BANK_0_PAGE_0
...
}
}
Listing 9. Splitting memory into pages
Pages for ADSP-BF561 Processors

ADSP-BF561 processors have a 32-bit memory interface. When using the 16-bit interface, addressing is
the same as for ADSP-BF53x processors. The 32-bit address mapping scheme of the ADSP-BF561
processor is defined in Figure 45.
Figure 45. ADSP-BF561 memory mapping scheme
Table 3 shows page size for 32-bit EBIU with regard to the EBCAW bits.

a
EBCAW Page bytes in Hex
8 bits 0x400
9 bits 0x800
10 bits 0x1000
11 bits 0x2000
Table 3. Page sizes for 32-bit EBIU
What’s the advantage of accessing the pages separately? If we can ensure that we stay within a page, no
additional precharge and activate commands are needed, saving time.
Figure 46 shows a peripheral DMA approach to avoid page switches. The DMA writes the incoming data
to pages to different banks.
Figure 46. Open page DMA approach
6.3 SDRAM Performance Items for Core Accesses

Core accesses are the most performance-critical SDRAM accesses. So, if you do not use data cache for
any reason, you must have a strategy to organize accesses to the SDRAM. This section describes the
reason for this bottleneck and how to handle it.
Buffers are used to handle the data transfer between the system clock domain and the core clock domain.
These buffers are organized as a state machine, which enables you to change the core and system clock
without concern for internal processes. Depended on the core/system clock ratio, wait states are
introduced to organize reads and writes by the core to and from the system domain.
One approach to solving this problem is to work with DMA transfers. Therefore, we are using DMA to
transfer data that we want to process into internal memory. DMA has an internal FIFO buffer structure
and can read from SDRAM without additional time penalties. Software planning and design is essential,
especially when working with very large (size of the array >> internal data memory) multi-dimensional
arrays. In this case, your algorithm must access this array in different manners. You can use DMA to
order the array optimally for your algorithm.

a
6.3.1 Code Overlays
A disadvantage of using cache is that the cache expects a program flow that may not fit to your program;
thus, it results often in cache misses. Another approach is to load the necessary code into internal memory
by DMA transfer; the code will be organized by an intelligent overlay manager that can predict which
piece of code is needed next. Whether there are any performance improvements by using an overlay
manager depends very much on your system design and on program flow. For most applications, an
optimization of the cache will be a better and easier approach to improve system performance. The first
challenge is to identify the modules of your program that are relatively independent and do not need to
call each other directly. Separate the overlays from the program, and place their machine code in the
larger memory. Develop an overlay manager that organizes intelligent DMA transfers of the overlays
from the SDRAM to the internal memory.
6.4 SDRAM Performance Items When Using Cache
Code
Optimizing memory for cache access means reducing cache misses. We have to organize the code and
data in a way that minimizes cache misses. When optimizing the code for cache accesses, keep the code
straight as possible and declare functions that are not often used in the program code as inline. This keeps
the code compact and minimizes the number of cache misses. Functions that are often called should be
placed into internal memory, if possible.
Data
Look at the algorithm and try to find a way to perform sequential data accesses. An example for a Fast
Fourier Transformation (FFT) that accesses the data sequentially is shown in Writing Efficient Floating-
Point FFTs for ADSP-TS201 TigerSHARC® Processors (EE-218)[8]. In C, a multi-dimensional array has
the following order in memory (here a 3-D array):
A 0,0,0 , A 0,0,1 , A 0,0,2 , A 0,1,0 , A 0,1,1 , A 0,1,2 , A 0,2,0 , A 0,2,1 , A 0,2,´2 , A1,0,0 , A1,0,1 , A1,0,2 , A1,1,0 , A1,1,1 , A1,1,2 ,
A 1,2,0 , A 1,2,1 , A 1,2,´2 , A 2,0,0 , A 2,0,1 , A 2,0,2 , A 2,1,0 , A 2,1,1 , A 2,1,2 , A 2,2,0 , A 2,2,1 , A 2,2,´2
Listing 10 shows the code to fill an array by the sequential access.

for (i=0;i<3; i++)
{
for (j=0;j<3; j++)
{
for (k=0;k<3; k++)
{ MyArray[i][j][k] = getValue();
}
}
};
Listing 10. Accessing an array optimal

a
7 Optimizing Power Consumption
Blackfin processors are used often in portable applications or low-power applications. In such
applications, it is essential to drive the power consumption as low as possible. The SDRAM boosts power
consumption by a high percentage. Therefore, the right choice, use of, and configuration of SDRAM are
fundamental to low-power design. This section provides an overview of ways to minimize power
consumption.
7.1 Introduction: Power-Consumption Figures

Although various standard symbols (Table 4) are used by device manufacturers to define power
consumption, the procedures used to measure these figures vary. Thus, there are differences in
interpreting these values.
Symbo Meaning
l
I CC1 Operating current in active mode
I CC2 Precharge standby current
I CC3 No operating/ standby current
I CC4 Operating current in burst mode / all banks

activated
I CC5 Auto-refresh current
I CC6 Self-refresh current
Table 4. Power consumption measurement symbols
7.2 Tips for Lowering SDRAM Power Consumption

Following are tips toward lowering SDRAM power consumption.
Use as less SDRAM as possible.
Use 1.8V or 2.5V mobile SDRAM (this is not possible for all Blackfin processors, see Brief
Introduction to SDRAM).
Lower the refresh rate in a low-temperature environment. The refresh rate of 64 ms is specified for the
worst-case scenario – high temperature. Thus, you still have room left to lower it when you are
operating in a standard environment.
Try to do data transfers between memory banks (not within memory banks).
Enable the self-refresh bit (SRFS) in the EBIU_SDGCTL register. In this mode, the power dissipation of
the SDRAM is at the lowest point.

a
7.3 Mobile SDRAM
Use mobile SDRAM or low-power SDRAM for embedded applications with high power-consumption
requirements. In these applications, the SDRAM is used very infrequently; thus, most of the time it is in
the idle state. Mobile SDRAM offers special modes that reduce the power consumption when the
SDRAM is in idle mode. Features that are supported by Blackfin processors are temperature-compensated
self-refresh (TCSR) and partial-array self-refresh. The temperature-compensated self-refresh feature
allows you to reduce the self-refresh frequency while it is in the idle state at temperatures below 45°C.
The leakage of the memory cells is very temperature dependent; when the temperature is high, the leakage
is higher than when the temperature is low. The self-refresh rate for standard SDRAM is set to a worst-
case value for the highest temperature, and the SDRAM is specified. At the Blackfin processor, you can
set the temperature via the TCSR bit of the EBIU_SDGCTL register. The value of TCSR indicates the
temperature border (for example, 45°C means you are operating below 45°C).
Many applications use most of the SDRAM devices only to buffer data arrays, which occurs after the
processing. For these applications, the partial-array self-refresh feature is a good approach to saving
power. This feature enables your application to select the memory banks that are to be refreshed during
idle mode. If you have any code on the SDRAM, place it at SDRAM bank 0 and bank 1; otherwise, it will
be lost.
For this application, use a low-voltage (1.8V or 2.5V) Blackfin processor.
7.4 Going into Hibernate and Recover

ADSP-BF537, ADSP-BF54x, and ADSP-BF52x processors preserve SDRAM content while the processor
is sent to hibernate mode. Therefore, the CKELOW bit in the VR_CTL register must be set to 1 to maintain the
CKE signal low during hibernate, which will prevent the SDRAM from losing data. The content of internal
memory and all registers (except the VR_CTL register) will be lost. Therefore, the program has to write all
the register settings and the internal memory content to SDRAM.
The following steps are used to go into hibernate mode and recover the data.
Step 1
Save all important registers to the SDRAM.
Save the important data of your internal memory to the SDRAM.
Ensure that the CKELOW bit is set in the VR_CTL register and the self-refresh bit in the SDGCTL register.
Send the processor into hibernate mode (Figure 47 and Listing 11).
Figure 47. Going into Hibernate mode

a
SaveTheMemory();
SaveTheRegister();
//Let's setup the RTC to wake up
SetupRTC();
IntMask = cli();
*pVR_CTL = ( WAKE | //wake by
CKELOW | //keeps the content of the SDRAM
CANWE | //ensures that the CAN RX can wake up the BF
(*pVR_CTL & ~FREQ)); // Send to hibernate
Listing 11. Going into Hibernate mode
Step 2
The processor is in hibernate mode (Figure 48). The content of all registers except VR_CTL are lost. The
data is stored in SDRAM.
Figure 48. Hibernate mode
Step 3
After waking up the processor from hibernate mode (Figure 49), the processor boots in the initialization
file. By checking the CKELOW bit, the processor determines whether it is coming from hibernate or from
reset. When the Blackfin processor is coming from reset, the processor continues the boot process;
otherwise, it calls a routine to restore the internal memory and the registers and then jumps to the
execution code.
Figure 49. Recovering from hibernate mode
7.5 Structuring Data for Low Power Consumption

Not only are the precharge and activate operations time consuming, they also have a massive impact on
power consumption. Therefore, keep the number of these operations as low as possible.
A precharge operation during the application will be executed at page breaks and SDRAM refreshes.
Therefore, organize the data in a way that keeps the number of page breaks as low as possible. Refer to
Increasing the SDRAM Performance of Your System for examples.

a
Appendices
Appendix A: Glossary
access time The time from the start of one device access to the time when the next
access can be started.
array Memory area for data storage. The array consists of rows and columns,
where each memory cell is located at an address where an intersection
occurs. Each bit in memory is found by its row and column coordinates
asynchronous A process where operations proceed independently.
auto precharge An SDRAM function that closes a page at the end of a burst operation.
auto refresh A mode where an internal oscillator establishes the refresh rate, and an
internal counter keeps track of the address to be refreshed.
bank A bank can mean the number of physical banks (same as rows) on the
SDRAM module. It can also mean the number of internal logical banks
(usually 4 banks nowadays) within an individual SDRAM device.
burst mode Bursting is a rapid transfer of data to a series of memory cell locations.
bypass capacitor A capacitor with the primary function of stabilizing a power supply voltage,
especially for an adjacent device or circuit
bus cycle A single transaction between a memory device and the system domain of
the Blackfin processor.
CAS Column address strobe. A control signal that latches a column address into
the SDRAM control register.
CAS-before-RAS (CBR) Column address strobe before row address strobe. CBR is a fast refresh
function that keeps track of the next row to be refreshed.
column Part of the memory array. A bit is stored where a row and column intersect.
crosstalk A signal induced in one wire or trace by current in another wire or trace
DDR Double data rate. The data is transferred on the rising and falling edge of the
clock. Since the address lines keep the same, data of sequential addresses
are transferred.
DQM Data mask signal used for masking during a write cycle. There is one DQM
signal per eight I/Os.
DRAM Dynamic Random Access Memory. A type of memory device usually used
for mass storage in computer systems. The term dynamic refers to the
constant refresh the memory must have to retain data.
EBIU External Bus Interface Unit. It provides mainly the synchronous external
memory interface to SDRAM which is compliant to the PC100 and PC133
standard and an asynchronous interface to SRAM, ROM, FIFO, flash
memory, and FPGA/ASIC designs.

a
EMC Compliance to rules and regulations controlling EMI
EMI Interference caused by electromagnetic radiation
external buffer An external buffer is needed to drive the SDRAM command, clock, clock
enable, and address pins, if they have a higher load than 50 pF. The
resulting capacitance is the number of input pins multiplied by the
capacitance of a SDRAM input pin (an input pin of the SDRAM has around
4.5 pF [consult your SDRAM data sheet]) plus the capacitance of the PCB
track.
FBBRW Fast back-to-back read-to-write. In the standard application, the write
command is delayed by one clock cycle after a read command is processed.
Fast-back-to-back-read-to-write enables to write directly after the read
command processing without the 1-cycle delay. Due to the fact that your
data bus has to switch quickly between the read and write data, this feature
is very dependent on the capacity of your data bus. This includes the
number of SDRAM chips and the design of the data bus circuit paths.
FPM Fast page mode. A common SDRAM data access scheme.
hibernate A special power mode of the Blackfin processor that provided the lowest
power dissipation. The I/O supply keeps established while the core is
powered down. The Blackfin processor can be awoken by several external
events.
interleaving The process of read/writing data alternately from two or more pages in the
SDRAM.
JEDEC Joint Electron Device Engineering Council
latency The length of time, usually expressed in clock cycles, from a request to read
a memory location and when the data is actually ready.
memory cycle time The time it takes for a complete memory operation (such as a read or write)
to take place.
microstrip A trace configuration where there is a reference plane on only one side of a
signal trace
OTP Memory One-Time-Programmable Memory. On-chip memory, which can be used
(among other things) to store setup values and keys for the Blackfin security
scheme.
page The number of bytes that can be accessed with one row address.
page mode An operation that takes place when RAS is taken logic low and a column
address is strobed in. The SDRAM remembers the last row address and
stays on that row and moves to the new column address.
PCB Printed circuit board

a
RAS Row address strobe. The control signal that latches the row address into the
SDRAM. It is used in conjunction with the column address to select an
individual memory location.
RAS to CAS delay The time between a row access strobe and a column address strobe.
read time The time required for data to appear at the output once the row and column
address become valid. Read time is also referred to as access time.
refresh A periodic restoration of an SDRAM cell charge needed to maintain data.
refresh cycle The time period in which one row of an SDRAM is refreshed
refresh period The minimum time that each row in the SDRAM must be refreshed
row Part of the memory array. A bit is stored where a row and column intersect.
SDR Single data rate. The data is transferred only once at one clock cycle. This
definition was introduced to differentiate this SDRAM from DDR.
SDRAM self-refresh In normal operating mode, the Blackfin processor will control the refresh of
the data cells by sending an auto-refresh command. But if the application
has a need to give up control over the SDRAM )for example, when the
processor is going into hibernate mode or in a multiprocessor application),
the SDRAM has to take over the responsibility of its data consistency.
Another case to send the SDRAM in self-refresh is reducing power
consumption. When the Blackfin processor is sending a self-refresh
command, the SDRAM will clock itself and will do self-refresh cycles. The
disadvantage is the delay when you want to access a SDRAM in self-
refresh.
strobe An input control signal that latches data synchronously into the SDRAM
stripline A circuit board configuration in which a signal trace is placed between two
reference planes
synchronous memory A memory device that has its signals synchronized to a reference clock.
termination One or more components used in conjunction with a transmission line to
control signal reflections
write time The time from which data is latched into the SDRAM until it is actually
stored in a memory location.

a
Appendix B: Code Examples, Schematics, and Excursus
Initialization Code (Chapter 2)

Listing 12 shows an example of initialization in assembly language.
P0.H = hi(EBIU_SDRRC);
P0.L = lo(EBIU_SDRRC);
R0 = 0x406 (z);
w[P0] = R0;
P0.H = hi(EBIU_SDBCTL);
P0.L = lo(EBIU_SDBCTL);
R0 = EBCAW_9 | //Page size 512
EBE; //SDRAM enable
w[P0] = R0;
P0.H = hi(EBIU_SDGCTL);
P0.L = lo(EBIU_SDGCTL);
R0.H= hi( ~CDDBG & // Control disable during bus grant off
SCTLE ); // SDRAM clock enable
R0.L= lo( ~CDDBG & // Control disable during bus grant off
SCTLE ) ; // SDRAM clock enable
[P0] = R0;
Listing 12. Initialization code example in assembly language

U1A
A01 J14
K14 U2
A02
L14
A03
J13 23 2
A04 A0 DQ0
K13 24 4
A05 A1 DQ1
L13 25 5
A06 A2 DQ2
K12 26 7
A07 A3 DQ3
L12 29 8
A08 A4 DQ4
M12 30 10
A09 A5 DQ5
M13 31 11
A10 A6 DQ6
M14 32 13
A11 A7 DQ7
N14 33 42
A12 A8 DQ8
N13 34 44
A13 A9 DQ9
N12 22 45
A14 A10 DQ10
M11 35 47
A15 A11 DQ11
N11 36 48
A16 A12 DQ12
P13 20 50
A17 BA0 DQ13
P12 21 51
A18 BA1 DQ14
P11 53
A19 DQ15
38
CLK
ABE0 H12
H13 17
ABE1 CAS
37 R1 R2 R3 R4 R5 R6 R7 R8
P4 15 CKE TBD TBD TBD TBD TBD TBD TBD TBD
D15 P5 39 DQML
D14 N5 18 DQMH
D13 M5 16 RAS R9 R10 R11 R12 R13 R14 R15 R16
D12 P6 19 WE TBD TBD TBD TBD TBD TBD TBD TBD
D11 N6 VDD CS
D10 M6 1
D09 P7 14 VDD
D08 N7 27 VDD
D07 M7 VDD 3 VDD
D06 P8 9 VDDQ
D05 N8 43 VDDQ
D04 M8 49 VDDQ
D03 P9 6 VDDQ
Blackfin® Processor and SDRAM Technology (EE-326)

D02 N9 12 VSSQ
D01 M9 46 VSSQ
D00 52 VSSQ
G12 VSSQ
AMS3 F13
AMS2 F14 MT48LC16M16/TSOP54
AMS1 E14
AMS0 G13
AOE G14
ARE H14
Schematics to Interface SDRAM to the Blackfin Processor (Chapter 4)
AWE E13
ARDY
B14
CLKOUT_SCLK
B13
SCKE E12
SA10 D13
SRAS C14
SCAS D12
SWE C13
SMS
D14 VDD
BR N10
BGH P10
BG R17 Title
Interf ace the Blackf in to SDRAM
10k
ADSP-BF537 Size Document Number Rev
A <Doc> 1.0
Date: Wednesday , September 13, 2006 Sheet 1 of 1

connection between SDRAM and the Blackfin processor (rather than as an approach to signal integrity).
Page 48 of 56
The next few pages show implementation examples. These schematics are given to illustrate the right
a
a
ADSP-BF561 in 16-Bit Mode
52
46 VSSQ
12 VSSQ
6 VSSQ MT48LC16M16/TSOP54
VDD 49 VSSQ
43 VDDQ
9 VDDQ
3 VDDQ
U13A VDD 27 VDDQ
14 VDD
C3 1 VDD
A25 A2 VDD
A24 D4 19
A23 B2 SWE 16 CS
A22 F5 SRAS 18 WE
A21 A3 ABE1SDQM1 39 RAS
A20 C4 ABE0SDQM0 15 DQMH
A19 B3 SCKE 37 DQML
A18 F6 SCAS 17 CKE
A17 B4 CAS
A16 E5 38
A15 A5 CLK 53 D15
A14 E6 21 DQ15 51 D14
A13 B5 20 BA1 DQ14 50 D13
A12 C6 A13 36 BA0 DQ13 48 D12
A11 A6 A12 35 A12 DQ12 47 D11
A10 D7 SA10 22 A11 DQ11 45 D10
A9 F11 A10 34 A10 DQ10 44 D9
A8 C14 A9 33 A9 DQ9 42 D8
A7 B14 A8 32 A8 DQ8 13 D7
A6 G10 A7 31 A7 DQ7 11 D6
A5 B15 A6 30 A6 DQ6 10 D5
A4 G11 A5 29 A5 DQ5 8 D4
A3 D13 A4 26 A4 DQ4 7 D3
A2 A3 25 A3 DQ3 5 D2
A15 A2 24 A2 DQ2 4 D1
ABE3SDQM3 A14 ABE3SDQM3 23 A1 DQ1 2 D0
ABE2SDQM2 B13 A0 DQ0
ABE1SDQM1 E11
ABE0SDQM0 U5
M15
D31 J12
D30 L16
D29 K12
D28 L15
D27 K13
D26 K15
D25 K14
D24 J16
D23 J13 MT48LC16M16/TSOP54
D22 J15
D21 H14 52
D20 H16 46 VSSQ
D19 H13 12 VSSQ
D18 H15 6 VSSQ
D17 H12 49 VSSQ
D16 VDD 43 VDDQ
G12 9 VDDQ
D15 G13 3 VDDQ
D14 G15 27 VDDQ
D13 F14 VDD 14 VDD
D12 F16 1 VDD
D11 F12 VDD
D10 F15 19
D9 F13 SWE 16 CS
D8 E15 SRAS 18 WE
D7 D16 ABE1SDQM139 RAS
D6 D15 ABE0SDQM015 DQMH
D5 E14 SCKE 37 DQML
D4 C16 SCAS 17 CKE
D3 E12 CAS
D2 C15 38
D1 B16 CLK 53 D15
D0 21 DQ15 51 D14
A7 20 BA1 DQ14 50 D13
AMS3 E7 A13 36 BA0 DQ13 48 D12
AMS2 B7 A12 35 A12 DQ12 47 D11
AMS1 C8 SA10 22 A11 DQ11 45 D10
AMS0 C7 A10 34 A10 DQ10 44 D9
AOE B8 A9 33 A9 DQ9 42 D8
ARE A8 A8 32 A8 DQ8 13 D7
AWE D9 A7 31 A7 DQ7 11 D6
ARDY A6 30 A6 DQ6 10 D5
A11 A5 29 A5 DQ5 8 D4
SCLK0_CLKOUT A12 A4 26 A4 DQ4 7 D3
SCLK1_CLKOUT B10 A3 25 A3 DQ3 5 D2
SCKE D11 A2 24 A2 DQ2 4 D1
SA10 C10 ABE3SDQM3 23 A1 DQ1 2 D0
SRAS D10 A0 DQ0
SCAS E10
SWE A10 U14
SMS3 C9
SMS2 B9
SMS1 E9
SMS0
R37
B12 VDD
BR A13
BG C12
BGH
10k
ADSP-BF561_BGA

a
ADSP-BF561 in 32-Bit Mode
52
46 VSSQ
12 VSSQ
6 VSSQ MT48LC16M16/TSOP54
VDD 49 VSSQ
43 VDDQ
9 VDDQ
3 VDDQ
U15A VDD 27 VDDQ
14 VDD
C3 1 VDD
A25 A2 VDD
A24 D4 19
A23 B2 SWE 16 CS
A22 F5 SRAS 18 WE
A21 A3 ABE1SDQM3 39 RAS
A20 C4 ABE0SDQM2 15 DQMH
A19 B3 SCKE 37 DQML
A18 F6 SCAS 17 CKE
A17 B4 To Blackfin CAS
A16 E5 To SDRAM 38
A15 A5 CLK 53 D31
A14 E6 21 DQ15 51 D30
A13 B5 20 BA1 DQ14 50 D29
A12 C6 A14 36 BA0 DQ13 48 D28
A11 A6 A13 35 A12 DQ12 47 D27
A10 D7 SA10 22 A11 DQ11 45 D26
A9 F11 A11 34 A10 DQ10 44 D25
A8 C14 A10 33 A9 DQ9 42 D24
A7 B14 A9 32 A8 DQ8 13 D23
A6 G10 A8 31 A7 DQ7 11 D22
A5 B15 A7 30 A6 DQ6 10 D21
A4 G11 A6 29 A5 DQ5 8 D20
A3 D13 A5 26 A4 DQ4 7 D19
A2 A4 25 A3 DQ3 5 D18
A15 A3 24 A2 DQ2 4 D17
ABE3SDQM3 A14 A2 23 A1 DQ1 2 D16
ABE2SDQM2 B13 To Blackfin A0 DQ0
ABE1SDQM1 E11
ABE0SDQM0 To SDRAM 2 U16
M15 To Blackfin
D31 J12
D30 L16
D29 K12
D28 L15
D27 K13
D26 K15
D25 K14
D24 J16
D23 J13 MT48LC16M16/TSOP54
D22 J15
D21 H14 52
D20 H16 46 VSSQ
D19 H13 12 VSSQ
D18 H15 6 VSSQ
D17 H12 49 VSSQ
D16 VDD 43 VDDQ
G12 To SDRAM 1 9 VDDQ
D15 G13 3 VDDQ
D14 G15 27 VDDQ
D13 F14 VDD 14 VDD
D12 F16 1 VDD
D11 F12 VDD
D10 F15 19
D9 F13 SWE 16 CS
D8 E15 SRAS 18 WE
D7 D16 ABE1SDQM1 39 RAS
D6 D15 ABE0SDQM0 15 DQMH
D5 E14 SCKE 37 DQML
D4 C16 SCAS 17 CKE
D3 E12 To Blackfin CAS
D2 C15 38
D1 B16 CLK 53 D15
D0 21 DQ15 51 D14
A7 20 BA1 DQ14 50 D13
AMS3 E7 A14 36 BA0 DQ13 48 D12
AMS2 B7 A13 35 A12 DQ12 47 D11
AMS1 C8 SA10 22 A11 DQ11 45 D10
AMS0 C7 A11 34 A10 DQ10 44 D9
AOE B8 A10 33 A9 DQ9 42 D8
ARE A8 A9 32 A8 DQ8 13 D7
AWE D9 A8 31 A7 DQ7 11 D6
ARDY A7 30 A6 DQ6 10 D5
A11 A6 29 A5 DQ5 8 D4
SCLK0_CLKOUT A12 To SDRAM A5 26 A4 DQ4 7 D3
SCLK1_CLKOUT B10 A4 25 A3 DQ3 5 D2
SCKE D11 A3 24 A2 DQ2 4 D1
SA10 C10 A2 23 A1 DQ1 2 D0
SRAS D10 To Blackfin A0 DQ0 To Blackfin
SCAS E10
SWE A10 U17
SMS3 C9
SMS2 B9
SMS1 E9
SMS0
R40
B12 VDD
BR A13
BG C12
BGH
10k
ADSP-BF561_BGA

a
Excursus: Calculating Z0 (Chapter 4)
This section describes the Telegrapher’s equation approach.
The SDRAM connection is a transmission line. This is the reason the telegrapher’s equations accurately
model the propagation of electrical currents and voltages along the structure. The trace is not an ideal
conductor, so we have to use an equivalent circuit diagram (Figure 50) that includes the influence of the
trace itself.
Input
L
Rs
C Rp
Figure 50. Equivalent circuit including the influence of the trace itself
But a trace consists not only of one of these structures. We can imagine a connection line (Figure 51) as a
nearly infinite series of them.
Input
L L
Rs Rs
C Rp C Rp
Figure 51. A connection line made up of infinite traces
The variable ZC is a function of the impedance and depends on the frequency. If you want to write it
mathematically correct, you have to write ZC(ω).
The variable Z0 is a single-valued constant showing the value of characteristic impedance at a particular
frequency ω0.
We simplify the model by defining the impedances Y and Z:
Z = jωL+Rs
Y = jωC+1/Rp
The resulting impedance is the sum of the impedance of Z and the impedance of Y and all the other stages
in parallel. We assume we have n elements each with the same Rp, Rs, C, and L.

a
~ Z 1
Zc = +
n 1 Y
~ +
Zc n
Y~
Multiply both sides by (1 + Zc )
n
~ Y~ Z Y~ ~
Z c (1 + Z c ) = (1 + Z c ) + Z c ⇔
n n n
~ 2 Z Z~
Zc = + Zc ⇔
Y n
~ Z Z~
Zc = + Zc
Y n
Assuming a transmission line consists of infinite elements with the length 0:
Z Z~ Z jωL + Rs
Z c = lim + Zc = =
n →∞ Y n Y jωΧ + 1/Rp
As you increase the frequency, the terms R and G may eventually be neglected as they are overwhelmed
by jωL and jωC, respectively, leading to a steady plateau in impedance. The fine balance between the
inductive impedance jωL and the capacitive admittance jωC holds the impedance constant at high
frequencies. This constant-impedance plateau greatly aids the design of high-speed digital circuits, as it
makes possible the termination of transmission lines with a single resistor. The value of characteristic
impedance at the plateau is called Z0.
L
Z 0 = lim ( Z C (ω )) =
ω →∞ C
Calculating the Inductance of a Microstrip Trace
It is strongly recommended to measure these values, but to estimate the value you can use the following
formulas:
⎛ 2π ⋅ h ⎞
L ≅ 5 ⋅ ln⎜ ⎟
⎝ w ⎠
whereby:
L is the inductance in nH/inch
h ís the height above the plane (mils)
w is the trace width (mils)
Calculating the Capacitance
A
C=ε 0 * ε r *
d

a
whereby:
A is the length times the width of the trace
d is the distance between ground and the trace
This must be calculated for every ground plane.
IPC’s and Douglas Brooks’ approach of Z0 (Chapter 4)
IPC and Douglas Brooks offer some equations for PCBs, which are very easy to use.
Microstrip Trace
87 ⎛ 5.98 ⋅ H ⎞
Z0 = ln⎜ ⎟
ε r + 1.41 ⎝ 0.8 ⋅ W + T ⎠
Embedded Microstrip Trace
60 ⎛ 5.98 ⋅ H ⎞
Z0 = ln⎜ ⎟
⎡ ( −1.55⋅ H 1)
⎤ ⎝ 0.8 ⋅ W + T ⎠
ε r ⎢1 − e H ⎥
⎣ ⎦
Stripline Trace
60 ⎛ 1.9 ⋅ (2 H + T ) ⎞
Z0 = ln⎜ ⎟
ε r ⎝ 0.8 ⋅ W + T ⎠
Asymmetric Stripline Trace
80 ⎛ 1.9 ⋅ (2 H + T ) ⎞⎛ H ⎞
Z0 = ln⎜ ⎟⎜1 − ⎟
ε r ⎝ 0.8 ⋅ W + T ⎠⎝ 4 ⋅ H 1 ⎠
whereby H1 > H

a
Dielectric Constants of Printed Circuit Boards (PCBs) (Chapter 4)
The tables are taken from A Survey and Tutorial of Dielectric Materials Used in the Manufacture of
Printed Circuit Boards[7].
Several Commonly Available Woven Glass Reinforces Laminates
List of Non-Woven or Very-Low Glass Content Laminate Materials

a
References
[1] The ABCs of SDRAM (EE-126), Rev. 1, March 2002. Analog Devices, Inc.
[2] System Optimization Techniques for Blackfin Processors (EE-324), Rev. 1, July 2007. Analog Devices, Inc.
[3] ADSP-BF52x Blackfin Processor Hardware Reference (Volume 1 of 2) Revision 0.31 (Preliminary). May, 2008.
Analog Devices, Inc.
[4] ADSP-BF52x Blackfin Processor Hardware Reference (Volume 2 of 2) Revision 0.3 (Preliminary). September, 2007.
[5] ADSP-BF533 Blackfin Processor Hardware Reference. Rev 3.2, July 2006. Analog Devices, Inc.
[6] ADSP-BF537 Blackfin Processor Hardware Reference. Rev 3.0, December 2007. Analog Devices, Inc.
[7] ADSP-BF561 Blackfin Processor Hardware Reference. Rev 1.1, February 2007. Analog Devices, Inc.
[8] ADSP-BF522/523/524/525/526/527 Blackfin Embedded Processor Preliminary Data Sheet. Rev PrE, August 2008.
[9] ADSP-BF512/ADSP-BF514/ADSP-BF516/ADSP-BF518 Blackfin Embedded Processor Preliminary Data Sheet.
Rev PrC, October 2008. Analog Devices, Inc.
[10] ADSP-BF531/ADSP-BF532/ADSP-BF533 Blackfin Embedded Processor Data Sheet. Rev F, 2008. Analog Devices, Inc.
[11] ADSP-BF534/ADSP-BF536/ADSP-BF537 Blackfin Embedded Processor Data Sheet. Rev E, March 2008.
[12] ADSP-BF538/ADSP-BF539 Blackfin Embedded Processor Data Sheet. Rev B, 2008. Analog Devices, Inc.
[13] ADSP-BF561 Blackfin Embedded Processor Data Sheet. Rev C, 2007. Analog Devices, Inc.
[14] A Survey and Tutorial of Dielectric Materials Used in the Manufacture of Printed Circuit Boards - By Lee W. Ritchey,
Speeding Edge, for publication in November 1999 issue of Circuitree magazine. Copyright held by Lee Ritchey of
Speeding Edge, September 1999.
[15] Writing Efficient Floating-Point FFTs for ADSP-TS201 TigerSHARC® Processors (EE-218), Rev. 2, March 2004
[16] Micron Technical Note 48-09 LVTTL DERATING FOR SDRAM SLEW RATE VIOLATIONS
[17] High Speed Digital Design – A Handbook of Black Magic by Howard W. Johnson and Martin Graham, 1993 PTR
Prentice Hall, ISBN 0-13-395724-1
[18] High-Speed Signal Propagation – Advanced Black Magic by Howard W. Johnson and Martin Graham, 2003, PTR
Prentice Hal, ISBN 0-13-084408-X
[19] EMV-Design Richtlinien by Bernd Föste and Stefan Öing, 2003 Franzis’ Verlag GmbH, ISBN 3-7723-5499-8
Readings
[20] ADSP-BF53x/ADSP-BF56x Blackfin Processor Programming Reference. Rev 1.2, February 2007. Analog Devices, Inc.

a
Document History
Revision Description
Rev 2 – December 11, 2008 Provides a more detailed description of the SDRAM initialization. Also, adds
by Fabian Plepp information on drive strength control for ADSP-BF52x devices. Incorporates
support for ADSP-BF51x processors.
Rev 1 – May 12, 2008 Initial public release. Adds Low Power section, OTP and ADSP-BF52x processors.
by Fabian Plepp
Rev 0 – August 21, 2006 Internal version (draft).

by Fabian Plepp

Engineer-to-Engineer Note EE-326: Blackfin® Processor and SDRAM Technology

Uploaded by

Copyright:

Available Formats

Engineer-to-Engineer Note EE-326: Blackfin® Processor and SDRAM Technology

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Engineer-to-Engineer Note EE-326: Blackfin® Processor and SDRAM Technology

Uploaded by

Copyright:

Available Formats

Engineer-to-Engineer Note EE-326

Blackfin® Processor and SDRAM Technology

Blackfin® Processor and SDRAM Technology (EE-326) Page 2 of 56

Blackfin® Processor and SDRAM Technology (EE-326) Page 3 of 56

Blackfin® Processor and SDRAM Technology (EE-326) Page 4 of 56

1.1 Basics of SDRAM

Blackfin® Processor and SDRAM Technology (EE-326) Page 5 of 56

1.2 SDRAM Parameters in Blackfin Registers

1.2.1 EBCAW (SDRAM External Bank Column Address Width)

Cmd ACTIV Pre ACTIV READ

Address Row Column

Figure 2. SDRAM Timing

Blackfin® Processor and SDRAM Technology (EE-326) Page 6 of 56

1.3 Multiprocessor Environment Options

Blackfin® Processor and SDRAM Technology (EE-326) Page 7 of 56

1.5 Options to Fit the SDRAM Timing

Blackfin® Processor and SDRAM Technology (EE-326) Page 8 of 56

Figure 4. PLL_CTL register (copied from Hardware Reference)

1.5.1 Blackfin Output / SDRAM Input Equation (Write)

Blackfin® Processor and SDRAM Technology (EE-326) Page 9 of 56

tOLZ t SSDAT tOH tOHZ

2.1 SDRAM Initialization Via an Emulator and VisualDSP++ .XML Files

Blackfin® Processor and SDRAM Technology (EE-326) Page 10 of 56

Figure 7. Target Options dialog box

Figure 8: A portion of the ADSP-BF5xx-proc.xml file (VisualDSP++ 4.5)

Blackfin® Processor and SDRAM Technology (EE-326) Page 11 of 56

<?xml version="1.0" standalone="yes"?>

<register name="EBIU_SDRRC" reset-value="0x03A0" core="Common" />

Listing 1. Example custom board reset settings (VisualDSP++ 5.0)

2.2 Initialization Using Memory-Mapped Registers

Blackfin® Processor and SDRAM Technology (EE-326) Page 12 of 56

Figure 9. Example: portion of a data sheet for mobile SDRAM

Figure 10: Address configuration specification in the data sheet

Blackfin® Processor and SDRAM Technology (EE-326) Page 13 of 56

Listing 2. Initialization via registers

Blackfin® Processor and SDRAM Technology (EE-326) Page 14 of 56

2.3 Initialization Using System Services

//set the twr to 2 sclk + no time

//set TRAS to 45ns

//Temperature Compensation at 85°C

Blackfin® Processor and SDRAM Technology (EE-326) Page 15 of 56

Listing 3.Initialization of the EBIU structure

Table 1. Overview of EBIU commands

Listing 4. Initialization of the EBIU

Blackfin® Processor and SDRAM Technology (EE-326) Page 16 of 56

2.4 SDRAM Initialization by the Values in the OTP Memory

Figure 11. Dummy write option

/* Declare local variable */

otp_write(0x18, OTP_LOWER_HALF, &Data);

/* Assign SDRAM values */

/* The low data byte is the EBIU_SDGCTL register */

Blackfin® Processor and SDRAM Technology (EE-326) Page 17 of 56

Data.h = 0x80000000 | // dummy write bit

Listing 5. Initialization via PBS

Figure 12. Specifying project load options

Blackfin® Processor and SDRAM Technology (EE-326) Page 18 of 56

//Setup the PLL Settings

//Wait the VR settings are done

//Setup wait cycles until PLL is set