TMS626162, TMS626812 16MBit Synchronous DRAMs Technical Reference 1996 MOS Memory Products Printed in U.S.A., April 1996 SMOU002 TMS626162, TMS626812 16M-Bit Synchronous DRAMs Technical Reference IMPORTANT NOTICE Texas Instruments (TI) reserves the right to make changes to its products or to discontinue any semiconductor product or service without notice, and advises its customers to obtain the latest version of relevant information to verify, before placing orders, that the information being relied on is current. TI warrants performance of its semiconductor products and related software to the specifications applicable at the time of sale in accordance with TI's standard warranty. Testing and other quality control techniques are utilized to the extent TI deems necessary to support this warranty. Specific testing of all parameters of each device is not necessarily performed, except those mandated by government requirements. Certain applications using semiconductor products may involve potential risks of death, personal injury, or severe property or environmental damage ("Critical Applications"). TI SEMICONDUCTOR PRODUCTS ARE NOT DESIGNED, INTENDED, AUTHORIZED, OR WARRANTED TO BE SUITABLE FOR USE IN LIFE-SUPPORT APPLICATIONS, DEVICES OR SYSTEMS OR OTHER CRITICAL APPLICATIONS. Inclusion of TI products in such applications is understood to be fully at the risk of the customer. Use of TI products in such applications requires the written approval of an appropriate TI officer. Questions concerning potential risk applications should be directed to TI through a local SC sales office. In order to minimize risks associated with the customer's applications, adequate design and operating safeguards should be provided by the customer to minimize inherent or procedural hazards. TI assumes no liability for applications assistance, customer product design, software performance, or infringement of patents or services described herein. Nor does TI warrant or represent that any license, either express or implied, is granted under any patent right, copyright, mask work right, or other intellectual property right of TI covering or relating to any combination, machine, or process in which such semiconductor products or services might be or are used. Copyright 1996, Texas Instruments Incorporated Preface Read This First About This Manual The purpose of this technical reference is to give you detailed operational information about Texas Instruments (TI) synchronous dynamic random-access memories (SDRAMs). It is intended to be used with the specific device data sheets (listed below), which contain the electrical characteristics, operating requirements and conditions, as well as timing and switching information. Notational Conventions This document uses the following conventions: - The TMS626162 and TMS626812 SDRAM devices are included by the term TMS626xx2. The TMS626xx2 devices are available as TMS626xx2-12 and TMS626xx2-15. Throughout this document, the device speed indicator is used as an abbreviation of the full device name and applies to the TMS626162 and the TMS626812 (for example, the - 15 specification allows read latency of one, two, or three cycles to be used). Related Documentation From Texas Instruments The following data sheets describe the Synchronous DRAM and related support tools. To obtain a copy of any of these TI documents, call the Texas Instruments Literature Response Center at (800) 477-8924. When ordering, please identify the data sheet by its title and literature number. TMS626162 Synchronous DRAM Data Sheet (literature number SMOS683B) describes the 512K-word by 16-bit by 2-bank SDRAM. TMS626812 Synchronous DRAM Data Sheet (literature number SMOS687) describes the 1M-word by 8-bit by 2-bank SDRAM. TI is a trademark of Texas Instruments Incorporated. Read This First iii Preface Other Related Documentation Configurations for Solid State Memories, JEDEC Standard No. 21-C, Release 4, Electronic Industries Association, November 1993. If You Need Assistance. . . iv If you want to. . . Do this. . . Order Texas Instruments documentation Call the TI Literature Response Center: (800) 477-8924 Ask questions about product operation or report suspected problems Call the Semiconductor Information Center: (214) 644-5580 Running Title--Attribute Reference Contents 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Synchronous DRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Functional Comparison: SDRAM Versus DRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Multiple Product Offerings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1 1-2 1-2 1-4 2 General Description of the 16M-Bit SDRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-1 2.1 Overview of the SDRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2 2.2 Burst Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3 2.3 Latency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-4 2.4 Two-Bank Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-5 2.4.1 Two-Bank Row-Access Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-5 2.4.2 Two-Bank Column-Access Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-5 2.5 Bank Deactivation (Precharge) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-6 2.6 Chip Select . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-6 2.7 Data/Output Mask . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-7 2.8 CLK Suspend/Power-Down Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-7 2.9 Mode-Register Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-8 2.10 Refresh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-9 2.10.1 CAS-Before-RAS (CBR) Refresh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-9 2.10.2 Self Refresh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-9 2.11 Interrupted Bursts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-10 2.12 Design Comparison: JEDEC-Standard Versus TI's SDRAM . . . . . . . . . . . . . . . . . . . . . 2-10 3 Detailed Operations on the 16M-Bit SDRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1 3.1 Power-Up Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2 3.2 Seamless Read With Bank Interleaving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-4 3.3 Read With Bank Interleaving - Random Row Address, Autodeactivate . . . . . . . . . . . . 3-9 3.4 Write With Bank Interleaving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-13 3.5 Write With Bank Interleaving - Random Row Address, Autodeactivate . . . . . . . . . . . . 3-17 3.6 Read Burst Interrupted by a READ Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-21 3.7 Read Burst Interrupted by a WRT Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-23 3.8 Read Burst Interrupted by a STOP Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-26 3.9 Read Burst Interrupted by a DEAC/DCAB Command . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-28 3.10 Write Burst Interrupted by a READ Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-30 3.11 Write Burst Interrupted by a WRT Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-32 3.12 Write Burst Interrupted by a STOP Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-34 v Contents 3.13 3.14 3.15 3.16 3.17 3.18 3.19 A vi Write Burst Interrupted by a DEAC/DCAB Command . . . . . . . . . . . . . . . . . . . . . . . . . . . Single-Bit Write . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Clock Suspend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Power-Down-Mode Entry -- Both Banks Active . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Power-Down-Mode Entry -- Both Banks Deactivated . . . . . . . . . . . . . . . . . . . . . . . . . . . Self-Refresh Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data-Masking Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-36 3-38 3-40 3-44 3-46 3-48 3-50 Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-1 Running Title--Attribute Reference Figures 2-1. 2-2. 3-1. 3-2. 3-3. 3-4. 3-5. 3-6. 3-7. 3-8. 3-9. 3-10. 3-11. 3-12. 3-13. 3-14. 3-15. 3-16. 3-17. 3-18. 3-19. 3-20. 3-21. 3-22. Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2 Mode-Register Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-8 Power-Up Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-3 Seamless Read With Bank Interleaving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-7 Read With Bank Interleaving - Random Row Address, Autodeactivate . . . . . . . . . . . . . . 3-11 Write With Bank Interleaving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-15 Write With Bank Interleaving - Random Row Address, Autodeactivate . . . . . . . . . . . . . . 3-19 Read Burst Interrupted by a READ Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-22 Read Burst Interrupted by a WRT Command in the Same Bank . . . . . . . . . . . . . . . . . . . . 3-24 Read Burst Interrupted by a WRT Command in a Different Bank . . . . . . . . . . . . . . . . . . . 3-25 Read Burst Interrupted by a STOP Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-27 Read Burst Interrupted by a DEAC Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-29 Write Burst Interrupted by a READ Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-31 Write Burst Interrupted by a WRT Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-33 Write Burst Interrupted by a STOP Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-35 Write Burst Interrupted by a DEAC Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-37 Single-Bit Write (Read Burst) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-39 Clock Suspend During a Write Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-42 Clock Suspend During a Read Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-43 Power-Down-Entry Operation, Both Banks Active . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-45 Power-Down-Entry Operation, Both Banks Deactivated . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-47 Self Refresh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-49 Data-Masking Operation (TMS626812) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-51 Data-Masking Operation With Upper- and Lower-Byte Control (TMS626162) . . . . . . . . 3-52 Contents vii Running Title--Attribute Reference Tables 1-1. 1-2. 1-3. 2-1. 2-2. 2-3. 2-4. viii Functional Differences Between the DRAM and the SDRAM . . . . . . . . . . . . . . . . . . . . . . . . 1-2 Raw Data Throughput . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3 TMS626xx2 Product Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-4 Bit-Burst Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3 Read Latencies for Different Operating Frequencies and Clock Speeds . . . . . . . . . . . . . . 2-4 Comparison of JEDEC-Standard Features With Features on TI's TMS626162 and TMS626812 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-10 Comparison of Pipeline and Prefetch Versus JEDEC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-11 Chapter 1 Introduction This chapter introduces the synchronous dynamic random-access memory (SDRAM) and compares it with the standard DRAM. Topic Page 1.1 Synchronous DRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2 1.2 Functional Comparison: SDRAM Versus DRAM . . . . . . . . . . . . . . . . . 1-2 1.3 Multiple Product Offerings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-4 1-1 Synchronous DRAM / Functional Comparison: SDRAM Versus DRAM 1.1 Synchronous DRAM As the speed of processors and memory buses continues to increase, the speed of the standard DRAM (dynamic random-access memory) becomes increasingly inadequate. In order to improve the overall system performance, the DRAM operations have been synchronized to the system clock, creating a synchronous DRAM (SDRAM). This document, in conjunction with the Texas Instruments TMS626xx2 16M-bit SDRAM data sheets (see related documents in the preface), helps you use SDRAMs efficiently. 1.2 Functional Comparison: SDRAM Versus DRAM Some major differences between the SDRAM and the DRAM make the SDRAM easier to control than the DRAM. There are also functional differences, some of which are summarized in Table 1-1. Table 1-1. Functional Differences Between the DRAM and the SDRAM DRAM SDRAM No system clock Runs off system clock Level RAS control Pulsed RAS control 1-bank operation 2 banks for on-chip interleaving 1 transfer per column address Burst of 1, 2, 4, 8 or 256 transfers per column address Read latency is nonprogrammable Read latency is programmable A DRAM is an asynchronous device. Systems using DRAMs are required to incorporate wait states that match the performance specifications of the DRAM. Timing of commands is dependent upon the speed of the DRAM and not necessarily upon system speeds. To facilitate better interaction between the SDRAM and the system, all commands are referenced to the system clock; therefore, the wait-state times needed to match the asynchronous DRAM timing with the system clock are avoided. To keep a DRAM row active, it is necessary to provide circuitry to hold RAS low. Precharge of the row is initiated by bringing RAS high. The SDRAM accepts commands on the rising edge of the system clock. To activate a row, RAS needs to be held low only for the setup and hold times relative to that clock edge. To deactivate the row, a deactivate command is given on a rising clock edge. The deactivate command initiates the precharge of the row. The pulsed RAS allows the SDRAM to have two independent memory banks on the device with only one RAS terminal. TMS626xx2 devices are compatible with the low-voltage TTL (LVTTL) interface. 1-2 Synchronous DRAM / Functional Comparison: SDRAM Versus DRAM The two independent banks allow each SDRAM to have two different rows active at the same time. This allows the reading or writing of data to one bank while the other is being readied. The delay normally associated with precharging and activating a row can be hidden by interleaving the bank accesses. The SDRAM can achieve a greater data throughput than a DRAM (see Table 1-2). One reason for this is the bursting capability of the SDRAM. The SDRAM can burst a series of addresses (1, 2, 4, 8, or 256) based on a given starting address. The length of the burst can be programmed by the system. Unlike conventional DRAMs, which require that each address fetch be initiated by a column address sent to the chip, up to 256 consecutive addresses can be fetched internally before another column address has to be sent, unlike conventional DRAMs, which require that each address fetch be initiated by a column address sent to the chip. This allows consecutive addresses to be transferred sequentially in a very short period of time after the initial column address has been sent. This means that any time data is being transferred in sequential blocks, the SDRAM performs at a much faster rate than a conventional DRAM. The SDRAM also uses a wider internal data bus to achieve the greater data rate. The amount of time needed to have valid data output from a DRAM depends on when the column-address command is given. This causes the system control to be more complicated and can cause delay to accommodate the various timings. The SDRAM has the delay between column entry and data valid programmed by the system; this is known as the read latency. Read latency is programmed to be from one to three clock cycles. The read latency used for a given system depends on the frequency of the system clock. Table 1-2. Raw Data Throughput Bytes Per Transfer SDRAM Time Delay (ns) DRAM Time Delays (ns) 2 70 60 4 82 100 8 106 180 16 154 340 32 250 660 64 442 1300 128 826 2580 256 1594 5140 Introduction 1-3 Multiple Product Offerings 1.3 Multiple Product Offerings The shaded area in Table 1-3 indicates that this technical reference is applicable to these part numbers. Table 1-3. TMS626xx2 Product Information Organization Part Number Supply Voltage 4Mx4 TMS626402 2Mx8 TMS626802 1Mx16 2Mx8 Architecture Refresh Banks Maximum Speed 3.3 V 2-bit prefetch 4K 2 100 MHz 3.3 V 2-bit prefetch 4K 2 100 MHz TMS626162 3.3 V Pipeline 4K 2 83 MHz TMS626812 3.3 V Pipeline 4K 2 83 MHz NOTE: Thin small-outline package (TSOP) 1-4 Device Description Chapter 2 General Description of the 16M-Bit SDRAM This chapter contains an overview of the SDRAM and its functions. Topic Page 2.1 Overview of the SDRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2 2.2 Burst Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3 2.3 Latency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-4 2.4 Two-Bank Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-5 2.5 Bank Deactivation (Precharge) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-6 2.6 Chip Select . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-6 2.7 Data / Output Mask . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-7 2.8 CLK Suspend / Power-Down Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-7 2.9 Mode-Register Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-8 2.10 Refresh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-9 2.11 Interrupted Bursts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-10 2.12 Design Comparison: JEDEC-Standard Versus TI's SDRAM . . . . . . 2-10 2-1 Overview of the SDRAM 2.1 Overview of the SDRAM The Texas Instruments SDRAM devices are high-speed 16 777 216-bit synchronous dynamic random-access memories. All inputs and outputs are synchronized with the system clock (CLK) input to simplify system design with high-speed microprocessors. All inputs of the SDRAM are latched on the rising edge of CLK. The SDRAM outputs also are referenced to the rising edge of CLK. The SDRAM array is divided into two banks, which are accessed independently. Either bank must be activated before it can be accessed (that is, read from or written to). Refresh cycles refresh both banks alternately. Figure 2-1 is the functional block diagram for the SDRAM. Figure 2-1. Block Diagram CLK AND Array Bank T CKE CS DQ Buffer DQM RAS 8 or 16 DQs Control Array Bank B CAS W A0 - A11 12 Mode Register DQML and DQMU for the TMS626162 SDRAM The following basic commands or functions control most operations of the SDRAM: 2-2 Row-address entry/bank activate Column-address entry/write Column-address entry/read Bank deactivate CAS-before-RAS (CBR) refresh Self refresh Overview of the SDRAM / Burst Sequence In addition to these basic commands or functions, three methods control the operation of the SDRAM: - CS to select / deselect the chip DQM to enable/mask the DQ signals on a cycle-by-cycle basis. For the x16 SDRAM, upper- and lower-byte control are used to enable and mask DQ signals. CKE to suspend (or gate) the CLK input 2.2 Burst Sequence All data for the SDRAM is written or read in burst fashion. Given a single starting address, the SDRAM internally accesses a sequence of locations based on that starting address. Some of the subsequent accesses can be at preceding column addresses and some can be at succeeding column addresses, depending on the starting address entered. The sequence can be programmed to follow either a serial- or an interleaved-burst sequence. The length of the burst sequence is user-programmable to be any of 1, 2, 4, 8, or 256 accesses. The TMS626162 and TMS626182 SDRAMs have a single-bit-write option. This allows the user to program the SDRAM to read a burst of 1, 2, 4, 8, or 256 bits, but write a burst of only 1 bit. In the case where this option is not used, the read and write burst lengths are equal. Table 2-1. Bit-Burst Sequences Internal Column Address A1 A0 Decimal Burst Type Serial Interleaved Binary Start 2nd 3rd 4th Start 2nd 3rd 4th 0 1 2 3 00 01 10 11 1 2 3 0 01 10 11 00 2 3 0 1 10 11 00 01 3 0 1 2 11 00 01 10 0 1 2 3 00 01 10 11 1 0 3 2 01 00 11 10 2 3 0 1 10 11 00 01 3 2 1 0 11 10 01 00 General Description of the 16M-Bit SDRAM 2-3 Latency 2.3 Latency The delay between the read command (READ) and the first output burst is known as the read latency (also referred to as CAS latency). The first data output cycle of a read burst can be programmed to occur one, two, or three CLK cycles after the read command. Minimum read latency is based on the particular maximum frequency rating of the SDRAM. This maximum frequency rating is provided in the data sheet as the minimum clock period, tCK. Table 2-2 summarizes read-latency requirements for the two speeds of the TMS626162 and TMS626812 SDRAMs. Table 2-2. Read Latencies for Different Operating Frequencies and Clock Speeds Device Speed Indicator - 12 Read Latency (Cycles) - 15 Clock Clock Speed Frequency Speed Frequency (ns) (MHz) (ns) (MHz) 36 28 40 25 2 18 56 20 50 3 12 83 15 66 1 For example, a - 15 device can operate over three ranges of frequencies ( 25 MHz, 25 - 50 MHz, and 50 - 66 MHz), but the allowable read latency in each case is different. At frequencies of 25 MHz and below, the - 15 specification allows a read latency of one, two, or three cycles to be used. At frequencies between 25 MHz and 50 MHz, the read latency can be either two or three cycles. Finally, at frequencies between 50 MHz and 66 MHz, the read latency must be three cycles. The programmable read-latency feature is provided to allow efficient use of the SDRAM over a wide range of clock frequencies. There is no latency for data-in cycles (write latency). The first data-in cycle of a write burst is entered at the same rising edge of CLK on which the write command (WRT) is entered. Note that the write latency is fixed and is not determined by the contents of the mode register. 2-4 Two-Bank Operation 2.4 Two-Bank Operation The SDRAM contains two independent array banks that can be accessed individually or in an interleaved, or seamless, fashion. Each bank can have one row activated at any given time. This is achieved by executing a bank-activate command (ACTV) with the bank selected by the state of address terminal A11 and the specific row selected by the state of address terminals A0 - A10. Each bank must be deactivated before it can be activated again with a new row address. A bank can be deactivated automatically by using the READ-P or WRT-P commands, or deactivated by using the DEAC command during a READ or WRT operation. Both banks can be deactivated simultaneously by use of the DCAB command. The availability of two banks allows enhanced performance and a wider variety of choices of possible combinations and methods of data access from which to choose, based on system needs. 2.4.1 Two-Bank Row-Access Operation The two-bank design allows you to access information on random rows at a higher rate of operation than is possible with a standard DRAM. Accomplish this by activating one bank with a row address as described previously, then, while the data stream is being accessed to / from the bank, activate the second bank with another row address. When the data stream to / from the first bank is complete, the data stream to / from the second bank commences without interruption. After the second bank is activated, you can deactivate the first bank to allow the entry of new row address for the next round of accesses. Operation can continue in this interleaved "ping-pong" fashion. 2.4.2 Two-Bank Column-Access Operation The availability of two banks also allows you to access data between banks from random starting columns at a higher rate of operation. After activating each bank with an ACTV command, use A11 to alternate READ or WRT commands between the banks to provide gapless accesses at the CLK frequency, provided all specified timing requirements are met. General Description of the 16M-Bit SDRAM 2-5 Bank Deactivation (Precharge) / Chip Select 2.5 Bank Deactivation (Precharge) Both banks can be deactivated simultaneously, or placed in precharge, by use of the DCAB command. A single bank can be deactivated by use of the DEAC command. The DEAC command and the DCAB command are differentiated from one another by the use of the A10 terminal. When A10 is held high during a deactivate command, the DCAB command is executed and both banks are deactivated. When A10 is held low, the DEAC command is executed and the state of A11 determines which bank is deactivated. Each bank can also be deactivated automatically by using the autodeactivate read (READ-P) and autodeactivate write (WRT-P) commands. The READ-P and WRT-P commands are distinguished from the standard read and write operations by the state of A10. Autodeactivate commands are selected when A10 is held high at the entry of a read or write operation. 2.6 Chip Select The SDRAM features a chip-select input, CS, that can be used to select or deselect the SDRAM for command entry. This provides a means for using the SDRAM in memory systems that require multiple memory device decoding. Hold the CS input high on the rising edge of CLK to deselect the device. This is a DESL command and affects only the RAS, CAS, and W inputs. The device remains in its present state until CS is brought low and a valid command is input. Use of CS does not affect an access burst that is in progress. The DESL command is equivalent to the NOOP command and the two can be used interchangeably. The device can be selected or deselected on a cycle-by-cycle basis. 2-6 Data / Output Mask / CLK Suspend / Power-Down Mode 2.7 Data/Output Mask Masking of individual data cycles within a burst sequence is accomplished by use of the MASK command. During a write burst, if DQM (DQMU or DQML) is held high on the rising edge of CLK, then the incident (referenced to the same rising edge of CLK) input on the DQs is ignored. For a read burst, if DQM (DQMU or DQML) is held high on the rising edge of CLK, the output data on the DQs is referenced to the second rising edge of CLK and is placed in the high-impedance state. Therefore, the application of DQM (DQMU or DQML) to data-output cycles (read bursts) involves a latency of two CLK cycles, while the application of DQM (DQMU or DQML) to data-in cycles (write bursts) has no latency. Furthermore, the MASK command (or its opposite, the ENBL command) is performed on a cycle-by-cycle basis, allowing you to gate any individual data cycle, or multiple cycles, within either a read or a write burst sequence. 2.8 CLK Suspend/Power-Down Mode For normal device operation, CKE should be held high to enable CLK. If CKE is brought low during the execution of a read or write operation, the state of the DQ bus occurring at the immediate next rising edge of CLK is frozen and no further inputs are accepted until CKE is returned high. This is known as a CLKsuspend operation and it is executed with a HOLD command. The device resumes operation from the point at which it was placed in suspension, beginning with the second rising edge of CLK after CKE is returned high. The device enters power-down mode if CKE is brought low when no read or write command is in progress ( PDE command ). If both banks are deactivated when power-down mode is entered, the power consumption is reduced to the minimum. Power-down mode can be used during row-active periods or CASbefore-RAS refreshes to reduce input-buffer power. After power-down mode is entered, no further inputs are accepted until CKE is returned high. When power-down mode is exited, new commands can be entered on the first CLK edge after CKE is returned high, provided that the setup time for CKE (tCESP) is satisfied. If tCESP > tCK, then NOOP or DESL commands must be entered until tCESP is met. CLK must be active and stable (if CLK was turned off for power-down) before CKE is returned high. The TMS626162 SDRAM has DQ mask capability with upper- and lower-byte (DQMU and DQML) control. General Description of the 16M-Bit SDRAM 2-7 Mode-Register Set 2.9 Mode-Register Set The SDRAM contains a mode register that you can program the read latency, the burst type, and the burst length. This is accomplished by executing an MRS command with the informatin entered on the address lines A0 - A9. A logic 0 must be entered on A7 and A8, but A10 - A11 are don't care entries for the SDRAM. When A9 = 1, the write burst length is always 1. When A9 = 0, the write burst length is defined by A2 - A0. Figure 2-2 shows the valid combinations for a successful MRS command. Only valid addresses allow the mode register to be changed. Figure 2-2. Mode-Register Programming A11 A10 A9 Reserved A8 A7 0 0 Register bit A9 Write burst length 0 A2-A0 1 1 A6 A5 A4 A2 A1 A0 Burst Type 0 = Serial 1 = Interleave Register Bits A6 A5 A4 Read Latency 0 0 0 0 1 1 1 0 1 1 2 3 All other combinations are reserved. Refer to timing requirements for minimum valid read latencies based on maximum frequency rating. 2-8 A3 Register Bits A2 A1 A0 0 0 0 0 1 0 0 1 1 1 0 1 0 1 1 All other reserved. B rst Length Burst 1 2 4 8 256 combinations are Refresh 2.10 Refresh The SDRAM, as the name implies, is still a dynamic RAM and must be refreshed, just as all other dynamic RAMs must be. All rows of the SDRAM must be refreshed at intervals not exceeding the refresh-period specification, tREF , to ensure data retention. The SDRAM offers two commands to accomplish this task: REFR and SLFR. In addition to these commands, refresh can be accomplished manually by performing an ACTV operation in each of the 4 096 rows within the refresh period. Because the SDRAM is divided into two independent and equal banks, 2 048 of the read or write operations must be performed in each bank. The use of the REFR and SLFR commands is discussed in more detail in following subsections. 2.10.1 CAS-Before-RAS (CBR) Refresh Before performing a CAS-before-RAS refresh, both banks must be deactivated. The refresh address is generated internally such that after 4 096 REFR commands, both banks of the SDRAM have been refreshed. The external address and bank-select (A11) inputs are ignored. The execution of a REFR command automatically deactivates both banks upon completion of the internal CBR cycle. This allows consecutive REFR-only commands to be executed without any intervening DEAC commands. The REFR commands do not necessarily have to be consecutive, but 4 096 of them must be completed before the refresh interval tREF expires. 2.10.2 Self Refresh To enter self refresh, you must deactivate both banks of the SDRAM. Following this deactivation, execute a SLFR command. For proper entry of the SLFR command, bring CKE low for the same rising edge of CLK that RAS and CAS are brought low and W is brought high. The CKE input must remain low for the device to stay in the self-refresh mode. In the self-refresh mode, all refresh signals are generated internally, and all inputs except CKE are ignored. Power consumption is reduced to a minimum and data is retained by the device for an indefinite period as long as power is maintained. To exit the self-refresh mode, return CKE high. You can issue new commands after waiting for the time interval tRC. If CLK is made inactive during self refresh, it must be returned to an active and stable condition before CKE is brought high to exit self refresh. General Description of the 16M-Bit SDRAM 2-9 Interrupted Bursts / Design Comparison: JEDEC-Standard Versus TI's SDRAM 2.11 Interrupted Bursts A read or write operation can be interrupted before the burst sequence has been completed with no adverse performance by entering certain superseding commands, provided that all timing requirements are met. The command interrupting either a read or a write burst can be entered any number of cycles from the initial burst command as determined by the specification nCCD. Burst interruptions on autodeactivate read and write operations are not supported. 2.12 Design Comparison: JEDEC-Standard Versus TI's SDRAM The minimum feature set and the implementation-dependent parameters are summarized in Table 2-3 and Table 2-4. Table 2-3. Comparison of JEDEC-Standard Features With Features on TI's TMS626162 and TMS626812 TMS626162 TMS626812 JEDEC Yes (optional) 1 Yes (optional) 2 Yes (optional) 4 Yes Yes 8 Yes Yes 256 Yes (optional) Features Burst stop (STOP command) Burst lengths 2-10 Design Comparison: JEDEC-Standard Versus TI's SDRAM Table 2-4. Comparison of Pipeline and Prefetch Versus JEDEC Feature Latency Prefetch TMS626402 TMS626802 Pipeline TMS626162 TMS626812 JEDEC n/a n/a 2n 1n 2n 1 1 1 0 n/a Burst Length Column-to-Column Address Delay nCCD Precharge Timing nEP >1 tAPR 1 >1 tRWL tAPW 2 0 -1 n/a 3 -1 -1 n/a 1 0 0 0 2 -1 -1 -1 3 -2 -1 -1 1 tRP + tCK tRP n/a 2 tRP tRP - tCK n/a 3 tRP - tCK tRP - tCK n/a 1 tRP tRP tRP 2 tRP - tCK tRP - tCK tRP - tCK 3 tRP - 2tCK tRP - tCK tRP - tCK n/a x + tCK x n/a >1 n/a x x x 1 n/a y + tCK y n/a >1 n/a y y y 1 x and y values are dependent on the speed version of product selected. Refer to the data sheets for actual values. General Description of the 16M-Bit SDRAM 2-11 2-12 Chapter 3 Detailed Operations on the 16M-Bit SDRAM This chapter describes specific operations on the 16M-bit SDRAM. Each section discusses a different operation, but all operations use the following assumptions: - SDRAM device speed indicator = -15 (15-ns CLK cycle time) Read latency = 3 System-clock frequency = 66 MHz Burst length = 4 Topic Page 3.1 Power-Up Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-2 3.2 Seamless Read With Bank Interleaving . . . . . . . . . . . . . . . . . . . . . . . . . 3-4 3.3 Read With Bank Interleaving -- Random Row Address, Autodeactivate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-9 3.4 Write With Bank Interleaving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-13 3.5 Write With Bank Interleaving -- Random Row Address, Autodeactivate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-17 3.6 Read Burst Interrupted by a READ Command . . . . . . . . . . . . . . . . . . 3-21 3.7 Read Burst Interrupted by a WRT Command . . . . . . . . . . . . . . . . . . . 3-23 3.8 Read Burst Interrupted by a STOP Command . . . . . . . . . . . . . . . . . . 3-26 3.9 Read Burst Interrupted by a DEAC / DCAB Command . . . . . . . . . . . 3-28 3.10 Write Burst Interrupted by a READ Command . . . . . . . . . . . . . . . . . . 3-30 3.11 Write Burst Interrupted by a WRT Command . . . . . . . . . . . . . . . . . . . 3-32 3.12 Write Burst Interrupted by a STOP Command . . . . . . . . . . . . . . . . . . 3-34 3.13 Write Burst Interrupted by a DEAC / DCAB Command . . . . . . . . . . . 3-36 3.14 Single-Bit Write . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-38 3.15 Clock Suspend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-40 3.16 Power-Down-Mode Entry -- Both Banks Active . . . . . . . . . . . . . . . . . 3-44 3.17 Power-Down-Mode Entry -- Both Banks Deactivated . . . . . . . . . . . 3-46 3.18 Self-Refresh Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-48 3.19 Data-Masking Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-50 3-1 Power-Up Sequence 3.1 Power-Up Sequence The power-up sequence ensures that all the internal logic circuits for the SDRAM are initialized to the proper state. After the device is powered up to the full VCC level, all device inputs must be held stable at a valid logic 0 or logic 1 level for 200 s. The inputs should be set such that NOOP commands are entered during the power-up period. After waiting for 200 s, both banks must be deactivated. In the example shown in Figure 3-1, a DCAB command is used, however, two DEAC commands, one for each bank, also are acceptable. Next, eight REFR commands must be performed. The timing specification tRP must be satisfied before the first REFR command. For the -15 device, the clock-cycle equivalent of tRP is calculated by: t RP B t + 45 ns B 15 ns + 3 CK This calculation of 3 cycles agrees with Figure 3-1. The timing specification that determines the number of cycles that must occur between each successive REFR command is proportional to tRC. From the TMS626162 data sheet, tRC is 125 ns. A calculation similar to that used for tRP yields a result of 9 cycles. After the last REFR command and its associated cycle time have elapsed, the SDRAM is ready to be programmed. This is accomplished by setting the mode register for the desired read latency, burst length, and burst sequence. In the example, the read latency is 3, the burst length is 4, and the burst sequence is serial. The correct address to be entered on inputs A0 - A7 is 0x32 (hexadecimal). The MRS command with this address programs the SDRAM to operate in the desired mode. The final step in the power-up sequence is to wait for a time given by the specification nRSA. This specification determines the minimum number of cycles that must elapse between entry of the MRS command and entry of any other valid command. Once this number of cycles has elapsed, the SDRAM is ready for operation. 3-2 Figure 3-1. Power-Up Sequence tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 Bank T - - - - - - - - - - - - - - - - Bank B - - - DCAB - - REFR #1 - REFR #8 - - - MRS - ACTV - CLK tRC tRP tCK nRSA 200 ms VCC VCCQ tCS RAS CAS W R0 A10 A11 Mode A0 - A9 Hi - Z DQ R0 A9 = Don't Care A7, A8 = 0x0 A0-A6 = 0x32 DQM CS CKE Time Lapse 3-3 This time period contains 6 cycles. DQML and DQMU for the TMS626162 SDRAM Time Lapse Time Lapse Power-Up Sequence Detailed Operations on the 16M-Bit SDRAM tCH Seamless Read With Bank Interleaving 3.2 Seamless Read With Bank Interleaving Figure 3-2 illustrates a seamless read with two banks. It assumes that both banks are deactivated at the beginning of the time interval shown. The first step is to activate row R0 in bank B. This is accomplished by performing an ACTV command. According to the data sheet, a time interval tRCD is required before a READ command can be executed. Given a clock period of 15 ns and tRCD = 40 ns, it follows that a READ command must wait at least until the third cycle after the ACTV command before execution: +t Bt number of cycles + 40 B 15 + 2.67 number of cycles + 3 number of cycles (in whole numbers) RCD CK Data is available on the third cycle after the READ command. Data is valid after a time tAC measured from the rising-clock edge previous to the data-out cycle. After the read latency, the SDRAM outputs data from a number of locations equal to the burst length. In this example, it is four consecutive locations. Additional READ commands can be executed before the burst is completed. In order to achieve seamless data in single-bank operations, the next READ command must be executed "burst" number of cycles after the previous READ command cycle. For example, read commands with burst length = 8 should be 8 cycles apart for seamless operation. READ commands can be executed in this manner to achieve seamless data within the limits of tRAS max. In this particular example, the objective is to achieve seamless data using both banks instead of just one bank. In order for this to occur, you must activate bank T. With the SDRAM, it is possible to have a row in both banks active at the same time. However, data from only one bank can be output at any given time because of the shared data bus. Commands are directed to only one bank at a time using A11. Because you want data to begin on bank T as soon as the data on bank B is finished, a READ command must be issued "burst" cycles (four cycles) after the previous READ command cycle on bank B. In order to issue a READ command on the correct cycle, an ACTV command must be issued at least a time tRCD previous to that READ command cycle on bank T. This is possible because no other command is being issued on that cycle. Therefore, seamless data is achieved using two banks. In this example, bank B is deactivated to demonstrate the issues surrounding the DEAC command, not because it is necessary. Deactivating bank B also illustrates that it is possible to change rows and still maintain seamless data when using both banks. If you want to get data from a different row in the same 3-4 Seamless Read With Bank Interleaving bank, then you must wait a specified number of cycles, as shown by the following calculation: number of cycles delayed + (t B t ) ) (t B t ) ) read latency RP CK RCD CK where the quotients of the two divisions are rounded up to the next whole number. In this particular example, tRP is equivalent to four cycles, tRCD is equivalent to three cycles, and the read latency is three cycles. This means the delay would be ten cycles after the DEAC command was issued. For this reason, it is desirable to issue the DEAC command as soon as possible. The DEAC command is restricted by the parameter nEP. For this example, the maximum value allowed for nEP is -2 cycles; however, at -2 cycles, a READ command is being issued in bank T. Therefore, the earliest the DEAC command can be issued is one cycle before the last data-out cycle as shown in Figure 3-2. In order to achieve seamless data with the specified conditions, two or more bursts must be completed per row address. This is because of the delay required from the DEAC command to the first data output as discussed previously. If the burst length was set to eight, only one burst is required per row. Detailed Operations on the 16M-Bit SDRAM 3-5 Seamless Read With Bank Interleaving 3-6 Seamless Read With Bank Interleaving Figure 3-2. Seamless Read With Bank Interleaving tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Bank T - - - - - - - Bank B ACTV - - READ - - DOUT - ACTV DOUT D w/READ OUT Read Latency = 3 - - READ DOUT DOUT DOUT - - DOUT DOUT D w/DEAC OUT - DOUT D w/READ OUT - ACTV DOUT DOUT DOUT - - READ DOUT D w/DEAC OUT - - - - DOUT DOUT R1C3+2 R1C3+3 R2C4+0 R2C4+1 CLK tCK tRCD nEP = -1 tRP tRAS tRC tCS tCH RAS CAS W A10 R0 R1 R2 A11 A0 - A9 R0 C0 C1 R1 R0C0+1 R0C0+2 C2 C3 R2 R1C2+1 R1C2+2 C4 tAC R0C0+0 DQ R0C0+3 R0C1+0 R0C1+1 R0C1+2 R0C1+3 R1C2+0 R1C2+3 R1C3+0 R1C3+1 DQM CS CKE Notes: 1) Timing relationships shown are for bank B operations. 2) Seamless for two or more bursts per row address 3) DQM includes DQML and DQMU for the TMS626162 SDRAM. Detailed Operations on the 16M-Bit SDRAM 3-7 Seamless Read With Bank Interleaving 3-8 Read With Bank Interleaving -- Random Row Address, Autodeactivate 3.3 Read With Bank Interleaving - Random Row Address, Autodeactivate Figure 3-3 illustrates how the autodeactivate feature is used to maximize the data bandwidth of the SDRAM in situations when the row-address characteristics of an application are random. This allows you to deactivate automatically the current row after the data has been output. In this example, the READ-P command is entered for the first access in each row of each bank. The considerations for the placement of the ACTV and READ-P commands are the same as in Figure 3-2. The minimum cycle time for one bank is determined by the specification tAPR, which is a combination of the nonautodeactivate specifications nEP and tRP . The number of cycles for tAPR is calculated as follows: t APR + t ) nEP RP t CK + 45 ns ) (-2 15 ns) + 15 ns, or 1 cycle In the previous case, the minimum time from the last data out to the entry of the DEAC command was limited to -1 cycle because of a conflict with the entry of the READ command in the opposite bank. In this example, the conflict is eliminated with the use of the autodeactivate function, and the cycle time for a read operation is reduced by one clock cycle. The penalty for changing rows after only one access is two null cycles for every eight data-access cycles. Detailed Operations on the 16M-Bit SDRAM 3-9 Read With Bank Interleaving -- Random Row Address, Autodeactivate 3-10 Read With Bank Interleaving -- Random Row Address, Autodeactivate Figure 3-3. Read With Bank Interleaving - Random Row Address, Autodeactivate tCK = 15 ns Bank T Bank B Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 - - - - ACTV - - READ-P - - DOUT DOUT DOUT DOUT - ACTV - - READ-P - - DOUT DOUT DOUT ACTV - - READ-P - - DOUT DOUT DOUT DOUT - ACTV - - READ-P - - DOUT DOUT DOUT DOUT - - - R2C2+2 R2C2+3 R3C3+0 R3C3+1 R3C3+2 CLK tRCD tAPR tCK tRC tCS tCH RAS CAS W A10 R0 R1 R2 R3 A11 A0 - A9 R0 C0 R1 DQ C1 R0C0+0 R0C0+1 R2 R0C0+2 R0C0+3 R1C1+0 R1C1+1 C2 R1C1+2 R1C1+3 R3 C3 R2C2+0 R2C2+1 DQM CS CKE Timing relationships shown are for bank B operations. DQM includes DQML and DQMU for the TMS626162 SDRAM. Detailed Operations on the 16M-Bit SDRAM 3-11 Read With Bank Interleaving -- Random Row Address, Autodeactivate 3-12 Write With Bank Interleaving 3.4 Write With Bank Interleaving In using the bank interleaving feature with the WRT command, the timing considerations for bank activation are the same as for the read operation (Section 3.2). One difference between the read and write operations is in the latency associated with each operation. There is a delay from the time the READ command is entered to the time that data is available at the SDRAM output. This is not the case for a write operation. The applicable specification that determines the time from a WRT command to the time when data is first accepted is nWCD, which is 0. This means that the first bit of data is presented at the SDRAM inputs at the same time that the write command is entered. This is illustrated in Figure 3-4. Another timing parameter that is specific to the write operation is tRWL. This parameter defines the minimum time required between the last data in and entry of a bank-deactivation command. The number of cycles for tRWL in this example is calculated as follows: number of cycles + t B t + 30 ns B 15 ns, RWL CK or 2 cycles A bank can be deactivated before the last data has been output by the SDRAM (see Section 3.2). This is not the case for write operations. You must wait until the final bit of data has been entered and two additional cycles before deactivating a bank. Therefore, read operations are more effective than write operations at hiding the latency associated with the row-deactivate operation. Detailed Operations on the 16M-Bit SDRAM 3-13 Write With Bank Interleaving 3-14 Write With Bank Interleaving Figure 3-4. Write With Bank Interleaving tCK = 15 ns Bank T Bank B Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 - - - - - - - - ACTV - - WRT w/DIN DIN DIN DIN WRT w/DIN DIN DIN DIN - DEAC - - - ACTV - - WRT w/DIN DIN DIN DIN WRT w/DIN DIN DIN DIN - DEAC - - - ACTV - - WRT w/DIN DIN DIN DIN WRT w/DIN CLK tCK tRCD tRWL tRP tRAS tCS tCH RAS CAS W A10 R0 R2 R1 A11 A0 - A9 DQ R0 C0 R0C0+0 R0C0+1 R0C0+2 R0C0+3 C1 R1 R0C1+0 R0C1+1 C2 R0C1+2 R0C1+3 R1C2+0 R1C2+1 R1C2+2 R1C2+3 C3 R2 R1C3+0 R1C3+1 C4 R1C3+2 R1C3+3 R1C4+0 C5 R1C4+1 R1C4+2 R1C4+3 R2C5+0 DQM CS CKE Timing relationships shown are for bank B operations. DQM includes DQML and DQMU for the TMS626162 SDRAM. Detailed Operations on the 16M-Bit SDRAM 3-15 Write With Bank Interleaving 3-16 Write With Bank Interleaving -- Random Row Address, Autodeactivate 3.5 Write With Bank Interleaving - Random Row Address, Autodeactivate Similar to a read operation with bank interleaving and autodeactivate (Section 3.3), the write-with-autodeactivate feature can eliminate conflict between the DEAC command in one bank and other commands in the opposite bank. As Figure 3-5 shows, the placement of the ACTV and WRT-P commands is the same as in the previous examples. The new specification that must be considered when developing a write cycle that uses the WRT-P command is tAPW, the minimum time from the last data in to the activation of the next row in that bank. This specification is best understood by realizing that it is based on combining the two relevant specifications that were discussed in Section 3.4 (Write with Bank Interleaving), tRWL and tRP . This is precisely how the value for tAPW was derived. For our example, the minimum number of cycles for tAPW is calculated as follows: t APW in number of cycles + (t ) t ) B t + (30 ns ) 45 ns) B 15 ns + 5 cycles RWL RP CK The benefit of using the autodeactivate command can be seen by observing that the minimum number of cycles for tRWL and for tRP can be less than the number of cycles when calculated independently (for some speed configuration); therefore, the use of the WRT-P command can save one cycle in the total write cycle time for one bank. Even without the capability to precharge before the completion of a write burst, the data-in throughput is the same as in the read case in Section 3.2. This is because the write operation does not incur any latency from command entry to data-in entry, whereas the read operation does. Detailed Operations on the 16M-Bit SDRAM 3-17 Write With Bank Interleaving -- Random Row Address, Autodeactivate 3-18 Write With Bank Interleaving -- Random Row Address, Autodeactivate Figure 3-5. Write With Bank Interleaving - Random Row Address, Autodeactivate tCK = 15 ns Bank T Bank B Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 - - - - ACTV - - WRT-P w/DIN DIN DIN DIN - - - ACTV - - WRT-P w/DIN DIN DIN DIN - - - - ACTV - - WRT-P w/DIN ACTV - - WRT-P w/DIN DIN DIN DIN - DIN DIN DIN - - - - - R3C3+1 R3C3+2 R3C3+3 CLK tRCD tCK tAPW tCH tCS RAS CAS W A10 R0 R1 R2 R3 A11 A0 - A9 DQ R0 C0 R1 R0C0+0 R0C0+1 C1 R0C0+2 R0C0+3 R1C1+0 R2 R1C1+1 R1C1+2 R1C1+3 C2 R3 R2C2+0 R2C2+1 C3 R2C2+2 R2C2+3 R3C3+0 DQM CS CKE Timing relationships shown are for bank B operations. DQM includes DQML and DQMU for the TMS626162 SDRAM. Detailed Operations on the 16M-Bit SDRAM 3-19 Write With Bank Interleaving -- Random Row Address, Autodeactivate 3-20 Read Burst Interrupted by a READ Command 3.6 Read Burst Interrupted by a READ Command The interruption of read bursts is permitted as described in Section 2.11, Interrupted Bursts. The interrupting command can be entered any number of cycles following the entry of the read. There is one restriction on interrupting a read burst: The interruption of a READ-P command is not allowed. In the interruption of a read operation by another read operation, an interesting issue is the distinction between an interrupting read and a normal read. As discussed in Section 3.2 (Seamless Read With Bank Interleaving), a new READ command begins a normal read operation if it is entered "burst" cycles or more after the previous READ command. If, however, the new READ command is entered before "burst" cycles are complete, it interrupts the read operation in progress. Figure 3-6 shows an example of a read burst interrupted by a READ command. The first READ command is interrupted by another READ command in the same bank. The interrupted READ command continues normal execution until the read latency of the interrupting READ has been satisfied. With a read latency of three cycles, the first two portions of data are output by the first read. At this time, the interrupting read takes effect, but it is interrupted also by a READ command in the opposite bank. As in the previous case, only the first two portions of the second read operation are output. The third read operation interrupts the last two portions of the second read and outputs the full four portions without further interruption. As demonstrated by the second and third READ commands, the operation of the SDRAM during interrupted bursts is not dependent on the bank in which the interrupting command is entered. Detailed Operations on the 16M-Bit SDRAM 3-21 tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Bank T - - - - ACTV - Bank B ACTV - - READ - READ READ DOUT Read Latency = 3 - - DOUT DOUT DOUT DOUT - - - DOUT DOUT DOUT - DEAC - - - - R0C0+1 R0C1+0 R0C2+1 R0C2+2 R0C2+3 CLK nCCD = 2 nCCD = 1 RAS CAS W A10 R0 R1 A11 A0 - A9 R0 C0 R1 C1 DQ DQM CS CKE DQM includes DQML and DQMU for the TMS626162 SDRAM. C2 R0C0+0 R0C2+0 Read Burst Interrupted by a READ Command 3-22 Figure 3-6. Read Burst Interrupted by a READ Command Read Burst Interrupted by a WRT Command 3.7 Read Burst Interrupted by a WRT Command The READ command also can be interrupted by a write operation with two additional considerations. The first consideration is that there is no latency associated with a write operation, so the read operation is interrupted immediately by the entry of a WRT or WRT-P command. The second consideration is the possibility of data contention. Because of the immediacy of the write operation, it is possible for the SDRAM to output data on one rising edge of the clock, and for the system to send a write operation on the next rising edge of the clock. The resulting data contention causes erroneous data to be written to the SDRAM. This problem can be avoided by the use of the DQM input. In Figure 3-7, the first read in bank B is interrupted after nCCD = 4 cycles. The DQM input must be brought high nDOD + 1 cycles prior to the entry of the interrupting write operation, as specified in Table 7 of the TMS626162 data sheet (Read-Burst Interruption). With the DQM input at a high-logic level, the outputs of the SDRAM are turned off. The first READ causes data to be available after three cycles, but because of the high state of the DQM two cycles prior to the DOUT, no data is output. In Figure 3-8, a read operation is executed in bank T and is interrupted by a write in bank B after nCCD = 5 cycles. This case is a clearer demonstration of the need to avoid data contention. The first two bits of data are due to be clocked out on the two clock cycles before the interrupting write command is entered. The first bit does not cause contention, but the second one can. To mask this data from being output, the DQM input must be high nDOD + 1 cycles prior to the write operation. To mask the second and third data bits of the read burst, the DQM must be high two cycles prior to the second data-bit output and held high for the next cycle to mask the third bit. This provides one cycle to clear the DQ bus and avoids any data contention. Detailed Operations on the 16M-Bit SDRAM 3-23 Read Burst Interrupted by a WRT Command Figure 3-7. Read Burst Interrupted by a WRT Command in the Same Bank tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 Bank T - - - - - - - - - - - Bank B ACTV - - READ - - DOUT WRT w/DIN DIN DIN DIN R0C1+1 R0C1+2 R0C1+3 CLK nCCD = 4 nDOD+1 = 3 RAS CAS W A10 R0 A11 A0 - A9 R0 C0 DQ DQM CS CKE DQM includes DQML and DQMU for the TMS626162 SDRAM. 3-24 C1 R0C1+0 Read Burst Interrupted by a WRT Command Figure 3-8. Read Burst Interrupted by a WRT Command in a Different Bank tCK = 15 ns Burst Length = 4 Device Speed Indicator = -15 Read Latency = 3 Bank T READ - - DOUT DOUT DOUT - - - - Bank B - - - - - WRT w/DIN DIN DIN DIN - R0C3+1 R0C3+2 R0C3+3 CLK nCCD = 5 nDOD+1 = 3 RAS CAS W A10 A11 A0 - A9 DQ C2 C3 R1C2+1 R0C3+0 DQM CS CKE A previous ACTV instruction activated row R1 in bank T. A previous ACTV instruction activated row R0 in bank B. DQM includes DQML and DQMU for the TMS626162 SDRAM. Detailed Operations on the 16M-Bit SDRAM 3-25 Read Burst Interrupted by a STOP Command 3.8 Read Burst Interrupted by a STOP Command To understand the effects of the interrupting STOP command, it is helpful to refer to Table 7 in the TMS626162 data sheet (Read-Burst Interruption). There are three points that determine the effect of the STOP command. The first point is that there is a 2-cycle latency from the time the STOP command is entered to the time the SDRAM outputs are placed in the high-impedance state. The data-sheet specification that determines the time required to return the outputs to high impedance is tHZ, which in this case is 11 ns. It is possible, then, that in certain situations the STOP command has no effect on the data being output by the current read operation. Therefore, the effect of the interrupting STOP command depends on where the command is entered. In Figure 3-9, the first STOP command is input on cycle nCCD = 2 after the READ command; therefore, the first bit of data is output as normal, and the STOP command causes the SDRAM output to turn off on the next cycle. Only one bit of the 4-bit burst is output. The second STOP command is input on cycle nCCD = 3. The first two bits of data are output as normal and the third and fourth bits are stopped. The second point is that the STOP command affects only the output of data. The row in the bank in which the STOP was executed remains active. The last point involves when the next new command in that bank can be executed. The next read or write operation must wait for a minimum of two clock cycles after the entry of the STOP command. The new READ command following the first STOP command can be input after the minimum number of cycles because there is no data contention on the bus. This is not necessarily true for the new WRT command following the second STOP command. According to the specification, the output returns to the high-impedance state a maximum of 11 ns after the last rising clock edge, and the data to be written must be valid at least 2 ns before the next rising clock edge. For the example, this leaves 2 ns of margin to avoid data contention. Because of this, one additional delay cycle is shown to ensure that there is no bus contention due to the switch from a read cycle to a write cycle. The 2-cycle delay between entry of the STOP command and the next READ or WRT command means that there is a penalty associated with using the STOP command. This is an important point to keep in mind when determining the best way to prevent the SDRAM from sending or receiving data in increments that are less than the programmed-burst length. A MASK command can be more efficient in some cases. See Section 3.19 (Data-Masking Operation) for a full description of data-masking features of the SDRAM. 3-26 Figure 3-9. Read Burst Interrupted by a STOP Command tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Bank T - - - - - - - - - - Bank B ACTV - - READ - STOP DOUT READ - - - - DOUT D w/STOP OUT Read Latency = 3 - - - - - WRT w/DIN DIN DIN CLK nCCD = 3 1 cycle nCCD = 2 nBSD = 2 nBSD = 2 (min) RAS CAS A10 R0 A11 A0 - A9 R0 C0 C1 C2 tHZ tHZ DQ R0C0+1 DQM CS CKE DQM includes DQML and DQMU for the TMS626162 SDRAM. R0C1+0 R0C1+1 tDS R1C2+0 R1C2+1 R1C2+2 3-27 Read Burst Interrupted by a STOP Command Detailed Operations on the 16M-Bit SDRAM W Read Burst Interrupted by a DEAC / DCAB Command 3.9 Read Burst Interrupted by a DEAC/DCAB Command A read burst can be interrupted by a deactivate operation by using either the DCAB command or the DEAC command. A cycle-by-cycle analysis of the timing diagram in Figure 3-10 is used to study the relevant specifications. In the first case, the interrupting command, DEAC, is entered on the second cycle following the READ command. This DEAC command takes effect after nHZP cycles have elapsed. The value of nHZP is always equal to the read latency; in this example, three cycles. The SDRAM outputs two bits of the 4-bit burst before the interrupting DEAC command takes effect. It is also possible for a deactivate operation to have no effect on the interrupted read operation. This would have been the case in the example if the DEAC command had been entered after nCCD = 4 cycles. In either case, the selected bank is deactivated. The second instance of a DEAC command is included to illustrate the relationship between nCCD and nEP. In this case, the DEAC command is entered nCCD = 5 cycles after the READ command. The specification for nEP takes precedence over the specification for nCCD. Therefore, the DEAC command is not an interrupt, the read operation completes normally, and the selected bank is placed in the precharge state. The precharge time, tRP , must be satisfied before the next command can be input. The effect of the interrupting DEAC (or DCAB) is specified in Table 7 of the TMS626162 data sheet (Read-Burst Interruption). 3-28 Figure 3-10. Read Burst Interrupted by a DEAC Command tCK = 15 ns Bank T - Bank B READ Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 - - - - - - - - - - - - DEAC DOUT DOUT ACTV - - READ - - DOUT - - - DOUT DOUT w/DEAC DOUT CLK nEP = -1 nCCD = 2 tRP nHZP = 3 RAS CAS R1 A10 A11 A0 - A9 DQ C0 R1 R0C0+0 R0C0+1 DQM CS CKE A previous ACTV instruction activated row R0 in bank B. DQM includes DQML and DQMU for the TMS626162 SDRAM. C1 R1C1+0 R1C1+1 R1C1+2 R1C1+3 3-29 Read Burst Interrupted by a DEAC / DCAB Command Detailed Operations on the 16M-Bit SDRAM W Write Burst Interrupted by a READ Command 3.10 Write Burst Interrupted by a READ Command The write operation can be interrupted by the same commands that are allowed to interrupt read operations. The same restrictions that apply to read interrupts also apply for write interrupts. The interrupting command can be entered any number of clock cycles after the WRT command is entered; the WRT-P command cannot be interrupted. The write operation can be interrupted by a read operation. This case is relatively straightforward to implement (see Figure 3-11). The write operation is interrupted after nCCD = 2 cycles by a READ command in bank B. As described in Table 8 of the TMS626162 data sheet (Write-Burst Interruption), the data input on the previous cycle is written by the SDRAM. The READ command immediately supersedes the data input and the SDRAM accepts no further data. The read operation executes as described in Section 3.2 (Seamless Read With Bank Interleaving). 3-30 Write Burst Interrupted by a READ Command Figure 3-11. Write Burst Interrupted by a READ Command tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 Bank T - - - - - - - - - - - - Bank B ACTV - - WRT w/DIN DIN READ - - DOUT DOUT DOUT DOUT R0C1+0 R0C1+1 R0C1+2 R0C1+3 CLK nCCD = 2 RAS CAS W A10 R0 A11 A0-A9 DQ R0 C0 R0C0+0 C1 R0C0+1 Ignore Ignore DQM CS CKE DQM includes DQML and DQMU for the TMS626162 SDRAM. Detailed Operations on the 16M-Bit SDRAM 3-31 Write Burst Interrupted by a WRT Command 3.11 Write Burst Interrupted by a WRT Command The write operation also can be interrupted by another write operation as shown in Figure 3-12. The WRT command is interrupted after nCCD = 3 cycles by a WRT-P command. The interrupting-write operation immediately supersedes the write that is in process. The result is that only the first three bits of the WRT command are written to the SDRAM. The next four data bits are written into the SDRAM at the column address specified at the time the interrupting WRT-P command is entered. 3-32 Figure 3-12. Write Burst Interrupted by a WRT Command tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 Bank T - - - - - - - - - - - - - - - - Bank B ACTV - - WRT w/DIN DIN DIN WRT-P w/DIN DIN DIN DIN - - - - - ACTV CLK nCCD = 3 tAPW RAS CAS W R0 R1 A11 A0-A9 DQ R0 C0 R0C0+0 C1 R0C0+1 R0C0+2 R0C1+0 DQM CS CKE DQM includes DQML and DQMU for the TMS626162 SDRAM. R1 R0C1+1 R0C1+2 R0C1+3 3-33 Write Burst Interrupted by a WRT Command Detailed Operations on the 16M-Bit SDRAM A10 Write Burst Interrupted by a STOP Command 3.12 Write Burst Interrupted by a STOP Command The example in Figure 3-13 shows the result of a write operation that is interrupted by a STOP command. Unlike the case for a read operation interrupted by a STOP, the effect of the interrupting STOP command is immediate. The data on the cycle interrupted by the STOP command is ignored. The SDRAM ignores all remaining data-in cycles of the write burst. In the example, the STOP command is entered nCCD = 2 cycles after the WRT command. The first two data bits of the burst are written by the SDRAM normally. The last two data bits of the burst are ignored. Because the data is ignored as soon as the STOP command is recognized, it is possible for the system-memory controller to withhold the last two data bits without jeopardizing the data integrity of the SDRAM. The final consideration for the write interrupted by a STOP command is the same as the read case. The next command must not be entered until two clock cycles after the STOP command. 3-34 Write Burst Interrupted by a STOP Command Figure 3-13. Write Burst Interrupted by a STOP Command tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 Bank T - - - - - - - - - - - - Bank B ACTV - - WRT/ DIN DIN STOP - WRT/ DIN DIN DIN DIN - R0C1+1 R0C1+2 R0C1+3 CLK nBSD nCCD = 2 RAS CAS W A10 R0 A11 A0-A9 DQ R0 C0 R0C0+0 C1 R0C0+1 Ignore Ignore R0C1+0 DQM CS CKE DQM includes DQML and DQMU for the TMS626162 SDRAM. Detailed Operations on the 16M-Bit SDRAM 3-35 Write Burst Interrupted by a DEAC / DCAB Command 3.13 Write Burst Interrupted by a DEAC/DCAB Command The final operation that can be used to interrupt a WRT command is a bank deactivate. Either of the two bank-deactivate commands, DEAC or DCAB, can be used in this case. A DEAC command entered in bank T does not interrupt an operation in bank B. The DCAB command interrupts an operation ongoing in either bank. The implementation of a write operation interrupted by a DEAC is complicated by the write-recovery specification, tRWL. This is the specification that determines the time that must elapse from the last data-in cycle to the entry of a deactivate operation. As indicated in Table 7 of the TMS626162 data sheet (Write-Burst Interruption), the DQM input must be used to mask the data inputs so that this specification is not violated. The specific details are shown in Figure 3-14. The WRT command is interrupted after nCCD = 2 cycles by the DEAC command. The minimum number of cycles required to meet tRWL in this case is two cycles. The DQM input must be brought high one cycle before the DEAC command is entered in order to satisfy the conditions for interrupting a write operation. The result is that only one cycle of data is written to the SDRAM. 3-36 Write Burst Interrupted by a DEAC / DCAB Command Figure 3-14. Write Burst Interrupted by a DEAC Command tCK = 15 ns Device Speed Indicator = -15 Read Latency = 3 Burst Length = 4 Bank T - - - Bank B ACTV - - - - - WRT w/DIN Masked DEAC - - CLK nCCD = 2 tRWL RAS CAS W A10 R0 A11 A0-A9 DQ R0 C0 R0C0+0 Masked Ignore DQM CS CKE DQM includes DQML and DQMU for the TMS626162 SDRAM. Detailed Operations on the 16M-Bit SDRAM 3-37 Single-Bit Write 3.14 Single-Bit Write The read latency, burst length and burst type are defined by the user when programming the mode-register set. A burst length of 1, 2, 4, 8, or 256 bits is encoded on address lines A0 - A2 during execution of the MRS command. If A9 is held low during execution of the MRS command, then the write- and readburst lengths are equal. The TMS626162 and TMS626812 SDRAMs also have a single-bit-write option. If A9 is held high during execution of the MRS command, then the write-burst length is set to one and the read-burst length, specified by A0 - A2, can be 1, 2, 4, 8, or 256 bits. See Figure 3-15 for an example of a single-bit write. 3-38 Figure 3-15. Single-Bit Write (Read Burst) Burst Length = 4 Bank T Read Latency = 3 - - - - - - - - - - - - ACTV - - Read - - DOUT DOUT DOUT DOUT WRT-P/ DIN - Bank B CLK RAS CAS W A10 R0 A9 - A0 R0 C0 DQ C1 R0C0 + 0 R0C0 + 1 R0C0 + 2 R0C0 + 3 R0C1 + 0 R0C1 + 1 DQM CS CKE DQM includes DQML and DQMU for the TMS626162 SDRAM. 3-39 Single-Bit Write Detailed Operations on the 16M-Bit SDRAM A11 3.15 Clock Suspend This section and the next section (Power-Down-Mode Entry - Both Banks Active), discuss the clock-gating features of the clock enable terminal, CKE. The two uses of CKE are clock suspension and reduced-power consumption. In both cases, while CKE is held low, the SDRAM does not recognize any activity on CLK. The effect of bringing CKE low is dependent on the state of the SDRAM. The SDRAM is placed in clock-suspend mode if CKE is brought low while a read or write operation is in progress. If no read or write operation is in progress at the time CKE is brought low, the SDRAM enters the power-down mode. There are two keys to understanding the effect of a clock-suspend operation that is accomplished by the entry of the HOLD command. The first is that there is a latency associated with the execution of a HOLD command. This latency is determined by the timing specification, nCLE. This parameter defines the number of clock cycles that must elapse before the clock-suspend operation takes effect, and also the number of clock cycles that must elapse before a new command can be entered after the clock-suspend operation is exited. The value of nCLE is always one clock cycle. The second point is that the clock input is not recognized by the SDRAM while clock suspend is in effect. The use of the HOLD command is shown in Figure 3-16 for a write and in Figure 3-17 for a read. The clock is suspended twice during the write cycle of Figure 3-16. This is done to illustrate the effect of varying the length of a clock-suspend operation. CKE is first brought low before the rising clock edge of the second data bit of the 4-bit burst. The clock-suspend operation is valid after nCLE cycles, provided that the specification for CKE setup time, tCES = 2 ns, is satisfied. Because there is a one-cycle latency from the time CKE is brought low to the time the SDRAM responds to the clock-suspend request, the second data bit is written to the SDRAM. The clock input is inhibited for as many cycles as CKE is held low. In this case, CKE is held low for two cycles, so that the data presented to the SDRAM inputs is ignored on the two cycles after nCLE is satisfied. Because of the entry and exit latency associated with the HOLD command, the clock edges, and hence the data inputs, that are ignored do not correspond to the time CKE is held low. The offset is equal to nCLE. Therefore, the second clock suspend during the write operation is entered during the same cycle that the third data bit is written, and is effective during the next cycle. The fourth and final data bit is written on the following cycle. With a total of three clock-suspend cycles, the write operation for a burst length of four takes seven cycles to complete. Figure 3-17 shows that clock suspend has a similar effect on the read operation. The first clock suspend in the read cycle occurs during the same cycle that 3-40 Clock Suspend the first cycle in the read latency is counted. The second and third cycles of the read latency are delayed for two cycles because CKE is held low. The effective access time from the start of the read operation is extended to five cycles. The first two bits of the read operation are output normally. However, CKE is held low for another two cycles, causing the fourth data bit to remain on the SDRAM outputs for as many cycles. This example also points out the different effect that CKE has on the read and write operations. In the write operation, the data that is to be input to the SDRAM on the first cycle after nCLE is satisfied is not written. However, the data that is output by the SDRAM on the first cycle after nCLE is satisfied is output and held. This is because of the difference in how the SDRAM handles the read and write operation. In the discussion of the read latency in Section 3.2: the data actually is output by the SDRAM one clock cycle before it is valid. In the example, the clock-suspend command does not take effect until after the SDRAM already has sent the fourth data bit out. The next two clock cycles are ignored so that the SDRAM does not advance the next data bit until nCLE cycles after CKE is brought high. Data is written to the SDRAM in the same clock cycle it is presented. Therefore, the clock suspend affects the data input immediately. Detailed Operations on the 16M-Bit SDRAM 3-41 Clock Suspend Figure 3-16. Clock Suspend During a Write Cycle tCK = 15 ns Burst Length = 4 Device Speed Indicator = -15 Bank T - - - - ACTV Bank B ACTV - - WRT w/DIN DIN - - Clock Clock Ignored Ignored Read Latency = 3 - - - DIN Clock Ignored DIN R0C0+2 Ignore R0C0+3 CLK RAS CAS W A10 R0 R1 A11 A0-A9 DQ R0 C0 R1 R0C0+0 R0C0+1 Ignore DQM CS tCES tCEH tCES tCEH CKE nCLE DQM includes DQML and DQMU for the TMS626162 SDRAM. 3-42 Ignore Figure 3-17. Clock Suspend During a Read Cycle tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Bank T - - - - - Bank B ACTV - - READ - - - - Clock Clock Ignored Ignored - - - - Read Latency = 3 - - - Clock Clock D DOUT DOUT DOUT Ignored Ignored OUT CLK tCS RAS CAS W A10 R0 A0-A9 C1 R0 R0C1+0 R0C1+1 R0C1+2 DQ R0C1+3 DQM CS tCEH nCLE tCES CKE tCEH DQM includes DQML and DQMU for the TMS626162 SDRAM. tCES 3-43 Clock Suspend Detailed Operations on the 16M-Bit SDRAM A11 Power-Down-Mode Entry -- Both Banks Active 3.16 Power-Down-Mode Entry -- Both Banks Active The second clock-gating function of the SDRAM provides a reduced-power operating mode. This mode is entered when CKE is brought low when a bank is active and no read or write operations are in progress. The effect of holding CKE low in a power-down operation is similar to the clock-suspend case in its effect on the clock input buffer. In both cases, the external-clock signal is ignored by the SDRAM. However, for a power-down operation, all of the input buffers of the SDRAM are turned off, including the clock-input buffer, resulting in significant power savings. The standby current consumed by the SDRAM is reduced by 14 mA when the device is placed in power-down mode. The example in Figure 3-18 shows a power-down entry and exit sequence for the case when both banks of the SDRAM are active. The power-down mode is entered, as opposed to the clock-suspend mode, because the last data output in the read burst was completed on the cycle previous to the entry of the PDE command and the timing parameter tCES = 2 ns was satisfied. Like the clock-suspend case, the power-down mode takes one cycle before taking effect. Therefore, the clock must remain valid for at least one cycle after the PDE command has been entered. Once the SDRAM has entered power-down mode, the external clock can be stopped for the duration of the power-down period. However, if the clock is turned off, it must be restarted and be stable before bringing CKE high. To exit the power-down mode, CKE must be brought high tCESP = 12 ns before the rising edge of the clock. If this time is met, a new command can be input to the SDRAM. In cases where the time tCESP is not met, a new command can be input on the following rising edge of the clock. 3-44 Figure 3-18. Power-Down-Entry Operation, Both Banks Active tCK = 15 ns Bank T - Device Speed Indicator = -15 Burst Length = 4 READ Latency = 3 ACTV - - READ - - DOUT DOUT DOUT DOUT - - - - - - - DOUT DOUT DOUT DOUT - - - - - PDE - - READ Bank B READ CLK tCK No READ(-P) or WRT(-P) in progress Enter power-down CLK must be active and stable before returning CKE high Exit power-down (new commands can be issued) tCH RAS W R1 A10 A11 A0- A9 DQ C0 R1 C1 C3 R0C0+0 R0C0+1 R0C0+2 R0C0+3 R1C1+0 R1C1+1 R1C1+2 R1C1+3 DQM CS CKE tCES tCEH Time Lapse 3-45 A previous ACTV instruction activated row R0 in bank B. DQM includes DQML and DQMU for the TMS626162 SDRAM. tCESP Power-Down-Mode Entry -- Both Banks Active Detailed Operations on the 16M-Bit SDRAM CAS Power-Down-Mode Entry -- Both Banks Deactivated 3.17 Power-Down-Mode Entry -- Both Banks Deactivated The last example of the SDRAM's power-down mode is for the case when both banks are inactive. As explained in the section "Power-Down-Mode Entry - Both Banks Active", the reduction in power consumption is different depending on the state of the SDRAM. As shown in Figure 3-19, an autodeactivate write operation is used to deactivate bank B. It is assumed that bank T is already inactive. Also, from the description of the PDE command given in the "Both Banks Active" section, there must be one NOOP or DESL cycle preceding a PDE command entry. Because of this one-cycle delay, the SDRAM has enough time to complete the write operation and turn off the internal row-address circuitry. Therefore, both banks are inactive at the time the power-down operation commences. The rest of the power-down operation continues as described previously. 3-46 Figure 3-19. Power-Down-Entry Operation, Both Banks Deactivated tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 Bank T - - - - - - - - - - - - - - Bank B ACTV - - WRT-P w/DIN DIN DIN DIN - PDE - - - ACTV - CLK tAPW tCS RAS CAS W R1 R0 A11 A0 - A9 DQ R0 R1 C0 R0C0+0 R0C0+1 R0C0+2 R0C0+3 DQM CS tCEH tCES CKE DQM includes DQML and DQMU for the TMS626162 SDRAM. tCESP 3-47 Power-Down-Mode Entry -- Both Banks Deactivated Detailed Operations on the 16M-Bit SDRAM A10 Self-Refresh Operation 3.18 Self-Refresh Operation In addition to the power-down feature, the self-refresh feature provides the capability to reduce the power consumption of the SDRAM during refresh operations. The conditions for proper execution of the SLFR command are similar to those of the power-down mode. Both modes of operation require that CKE be held low, but there are two additional constraints on the self-refresh operation. The first constraint is that both banks must be deactivated and fully precharged prior to the SLFR command entry. The second constraint is that the first command after the self-refresh exit must wait for a time specified by tRC before it can be entered. These constraints are illustrated in Figure 3-20. In this example, bank T is considered to be precharged fully. To meet the first constraint as described in the previous paragraph, entry of the SLFR command must wait until the time tAPW has expired. A valid SLFR command is entered on the cycle shown, provided that CKE is held high for a time determined by the hold-time specification tCEH and is brought low in time to meet the setuptime specification tCES. The device remains in the self-refresh mode as long as CKE is held low. As is the case during a power-down operation, CLK and all other inputs are ignored while the device is in the self-refresh mode. To exit the self-refresh mode, CKE must be brought high in time to meet the specification for tCESP . If the system clock is off, it must be returned to an active and stable state prior to bringing CKE high. To meet the second constraint described in the previous paragraph, entry of the next command must wait for a time determined by the specification tRC. After this time, a valid command can be entered. Upon exiting self refresh, you must begin the normal-refresh scheme immediately. If you are using a burst-refresh scheme, then 4 096 REFR commands must be executed before continuing with normal device operations. If a distributed-refresh scheme using CBR is employed (for example, two rows every 32 s), then you must perform the first set of refreshes before continuing with normal device operation. This ensures that the SDRAM is fully refreshed. 3-48 Figure 3-20. Self Refresh tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Read Latency = 3 Bank T - - - - - - - - - - - - - - - - Bank B ACTV - - WRT-P w/DIN DIN DIN DIN - - SLFR - - - - - REFR CLK tAPW tRC CLK is a Don't Care for this period tCS RAS CAS W R0 A11 A0 - A9 DQ R0 C0 R0C0+0 R0C0+1 R0C0+2 R0C0+3 DQM CS tCESP CKE Time Lapse tCEH tCES Time Lapse 3-49 Time Lapse DQM includes DQML and DQMU for the TMS626162 SDRAM. Self-Refresh Operation Detailed Operations on the 16M-Bit SDRAM A10 Data-Masking Operation 3.19 Data-Masking Operation The data-masking feature of the SDRAM provides another way to perform read or write operations that are shorter than the programmed-burst length. This feature is useful in situations where the majority of data accesses are four bits in duration but a small percentage of the accesses are only two bits long. A data-masking operation can be performed on both read and write cycles, and is effective on a cycle-by-cycle basis. This allows the user to mask any single data cycle in a burst sequence. The DQM input is used to signal whether the data in a given cycle is to be masked or enabled. The data to be read or written is enabled by keeping DQM at a logic-low level during the appropriate cycle. The data is masked when DQM is held high, and the data is enabled when DQM is held low. These operations are performed by the MASK and ENBL commands, respectively. For write operations, the state of the input, either masked or enabled, is determined by the logic level of DQM in that same cycle. In other words, there is no latency associated with DQM input for write operations. The TMS626162 allows the data-masking operation with upper- and lowerbyte control. The control signals are DQML (lower byte) and DQMU (upper byte). The upper / lower byte is masked when DQMU / DQML is held high and upper / lower byte is enabled when DQMU / DQML is held low. See Figure 3-22 for an example of the data-masking operation with upper- and lower-byte control. There is, however, a latency associated with the DQM for read operations. This delay is specified by the timing parameter nDOD and is always two cycles. Therefore, the state of the output during a read operation is determined by the logic level of the DQM input two cycles before the expected data output cycle. The reason for the two-cycle delay is related to the way the SDRAM outputs data. Since data is output by the rising-clock edge of the cycle before the data is to be valid, the MASK command must be recognized one cycle before that in order to stop the data from being output. The result is the two-cycle latency for the MASK and ENBL commands. The example in Figure 3-21 shows the timing relationships for both the mask and enable operations during write and read cycles. The first and last bits of both the read and write bursts are masked by bringing DQM high. As discussed previously, the data input on the write cycles when DQM is high is ignored by the SDRAM. For the read cycles, the outputs of the SDRAM are placed in the high-impedance state in the second cycle after DQM is brought high. 3-50 Figure 3-21. Data-Masking Operation (TMS626812) tCK = 15 ns Device Speed Indicator = -15 Burst Length = 4 Bank T - - - - - - Bank B ACTV - - WRT Masked DIN DIN - - Masked READ - - - - - Read Latency = 3 - Masked DOUT - - DOUT Masked CLK nDOD RAS nDOD tCH tCS CAS W R0 A11 A0 - A9 DQ R0 C0 Masked C1 R0C0+1 R0C0+2 Masked R0C1+1 R0C1+2 DQM CS CKE 3-51 Data-Masking Operation Detailed Operations on the 16M-Bit SDRAM A10 Bank T Bank B tCK = 15 ns - - - ACTV - - - READ Burst Length = 4 - - - - - ACTV - DEAC - - - - - - WRT - - - - - - R1C1 + 1 R1C1 + 2 R1C1 + 3 CLK DQ0DQ7 R1C1 DQ8DQ15 R0C0 R0C0 + 1 R0C0 + 2 R0C0 + 3 DQMU DQML RAS CAS W A10 R0 R1 A11 A0A9 CS CKE R0 C0 R1 C1 Read Latency = 3 - - DEAC - - - Data-Masking Operation 3-52 Figure 3-22. Data-Masking Operation With Upper- and Lower-Byte Control (TMS626162) Appendix AppendixAA Glossary A ACTV command: See bank activation. autodeactivate read (READ-P command): Banks are automatically deactivated at the end of the read access; a separate command does not have to be issued. autodeactivate write (WRT-P command): Banks are automatically deactivated at the end of the write access; a separate command does not have to be issued. B bank activation (ACTV command): Command sequence that causes one or both memory banks to be in the active state. bank deactivation ( DCAB, DEAC commands): Banks can be active only for a certain length of time. Bank deactivation is a command sequence that causes one or both banks to be in the inactive or precharge state. burst: The capability of a memory device to fetch multiple addresses given only the starting address. burst interrupt, burst interruption: A command issued after the beginning of a burst sequence but before the completion of the burst, which begins a new transaction without causing adverse effects. This interrupting command sequence should be entered only on even numbers of cycles from the initial burst command. burst length: The number of addresses that can be fetched internally by the SDRAM before the next column address is sent. burst sequence: Defines the amount and type of data the SDRAM can handle without requiring a secondary instruction. Burst sequences can be 8, 4, 2, or 1 unit in length and specify either serial or interleaved addressing. A-1 Glossary C CLK suspend: The state the SDRAM enters when CKE goes low during a read or write operation. Clocks are ignored and the operation in progress is put in hold. The SDRAM resumes operation from the point at which it was placed in suspension once CKE is returned high. If CLK suspend is entered while both banks are deactivated, the SDRAM enters the powerdown mode. D data masking input (DQM): Inhibits read / write for the cycle. Data is not output / input during cycle, rather in the high-impedance state. DCAB command: Deactivate both memory banks. See bank deactivation. DEAC command: Deactivate one memory bank. See bank deactivation. DESL command: Deselect command. The device remains in its current state or continues the active process for each clock cycle without need for a new command. If no process is active, the SDRAM is idle and banks are precharging. DESL is equivalent to the NOOP command. H HOLD command: Command that initiates CLK-suspend / power-down mode. I interrupted bursts: See burst interruption. interleaving: Process by which transfer operations can occur by switching back and forth between the two banks internal to the SDRAM. J JEDEC: Solid State Products Engineering Council ( formerly Joint Electron Device Engineering Council ) of the Electronic Industries Association ( EIA). This council operates under EIA administrative and legal procedures and publishes JEDEC standards and publications. This council also continuously develops and maintains these standards as required by the industry. A-2 Glossary L latency: Number of clock cycles until command takes effect. The state of the device changes after the latency period. M MASK command: Command that initiates data or output masking; see masking. masking: Process by which individual data cycles within a burst sequence can be ignored or disabled (placed in the high-impedance state). mode-register set: The command process by which the programmable features of the SDRAM are defined. Such features include serial or interleave burst type, defining system read latency and defining burst length. MRS command: Mode-register set command. See mode-register set. N nBSD: Number of cycles from entry of STOP command to entry of next command. nCCD: Column-to-column address delay, or initial-command-burst-to-interrupting-command delay ( in cycles ). nCLE: Number of cycles from either the HOLD command entry to the suspended CLK edge or from CLK-suspend exit to entry of any valid command. nEP: Number of cycles from final data out to bank deactivation. nDOD: Number of cycles from an ENBL or MASK command to output data valid. nHZP: Number of cycles from bank-deactivation ( DEAC or DCAB ) interrupt of a read burst to the data outputs going into the high-impedance state. NOOP command: Similar to DESL command. See DESL command. nRSA: The minimum number of cycles that must elapse between MRS command entry and entry of any other valid command. nWCD: The number of cycles from a valid WRT command to the first valid data at the inputs. Glossary A-3 Glossary P PDE command: See power-down enable command. precharge: The state of the memory bank when deactivated. ping-pong operation: Process of interleaving accesses in both banks by going back and forth between bank T and bank B. power-down enable command (PDE): Puts part into low power consumption mode, thereby saving stand-by power. power-down mode: The power-saving mode the SDRAM enters if CKE goes low while both banks are inactive. To ensure validity of data during power-down mode, execute the SLFR command concurrently with the PDE command to activate self refresh. R READ command: Initiates a read access to the activated bank of the SDRAM. read latency: The delay between the read command (READ) and the first output burst (also referred to as CAS latency). READ-P command: See autodeactivate read. REFR command: Initiates a CAS-before-RAS refresh of the SDRAM after both banks have been deactivated. RAS and CAS must be low and W must be high on the rising edge of CLK. refresh: DRAM or SDRAM operation by which data is retained in the memory. In the SDRAM, refresh can be accomplished by performing an activate/ deactivate command sequence to every row in both banks, by performing 4096 REFR commands, or by placing the device in self refresh. S SDRAM: Synchronous dynamic random-access memory. A memory device in which DRAM operations are synchronized to the system clock. seamless (gapless) operation: Process by which interleaved operations of the SDRAM tightly coincide such that data is transferred each clock cycle without a gap or interruption. This is the fast mode of operation the SDRAM can perform with potential rates of 100 MHz depending on the device speed indicator of the product selected. A-4 Glossary SLFR command: See self refresh. STOP command: Discontinues acceptance of data at the inputs during a write access, or completes only the read access in process, whichever is applicable. The bank remains active, but a new command must be entered to start any operation. self refresh: Type of memory refresh that requires CKE to be held low in addition to CAS and RAS being low and W being high for one cycle. Sustaining the self-refresh action for subsequent cycles requires only that CKE remain low. Data is retained indefinitely while power is maintained. Power consumption is reduced to a minimum in self refresh. W write latency: The delay between the write command ( WRT ) and the time data is accepted at the memory inputs. There is no write latency for the SDRAM. WRT command: Initiates a write access to the activated bank of the SDRAM. WRT-P command: See autodeactivate write. Glossary A-5 A-6 Glossary This template is for the "See" and "See also" references in your index. Since these entries do not have a page number associated with them, it's extremely difficult to locate one if you need to modify or delete it and you don't remember which chapter it's in. By using this template, you can alphabetize your entries according to the first letter of the first level entry. A B C D E F G H I J Glossary 1-1 Glossary K L M N O P Q R interrupting a read burst, S T U V 1-2 Glossary W X Y Z Glossary 1-3 Index Index A ACTV command 2-5, 2-9, 3-4, 3-9, 3-17 autodeactivate read (READ-P) 2-6, 3-9 to 3-11 autodeactivate write (WRT-P) 2-6, 3-17 to 3-19, 3-44 B bank activate command (ACTV) 2-5 bank activation 1-3, 2-5 bank deactivation (precharge) 1-2, 2-6 block diagram 2-2 burst interruptions 2-10 burst length 1-3 burst sequence 2-3 C CAS-before-RAS (CBR) refresh 2-9 chip select (CS) 2-6 CLK-suspend mode 2-7, 3-38 to 3-40 clock suspend 3-38 to 3-40 D data throughput 1-3 data-masking operation 3-48 data/output mask 2-7 DCAB command 2-5, 2-6, 3-2, 3-28, 3-36 DEAC command 2-5, 2-6, 2-9, 3-2, 3-4 to 3-8, 3-17 to 3-20, 3-28, 3-36 design comparison: JEDEC-standard versus TI's SDRAM 2-10 to 2-12 DESL command 2-6, 3-44 detailed operations on the 16M-bit SDRAM 3-1 to 3-50 differences between the SDRAM and the DRAM 1-2 DQM input 3-23 E ENBL command 2-7, 3-48 F functional comparison: SDRAM versus DRAM 1-2 to 1-4 G gapless (seamless) accesses 2-5, 3-5, 3-21 general description of 16M-bit SDRAM 2-1 to 2-12 H HOLD command 2-7, 3-38 I interleaving operation 1-2, 1-3, 2-3, 2-5, 3-4, 3-9, 3-13, 3-17 interrupted bursts 2-10 interrupting a read burst 3-21 to 3-30 See also read burst interrupting a write burst 3-30 to 3-37 See also write burst introduction 1-1 to 1-4 Index-1 Index J JEDEC-standard SDRAM features 2-10 prefetch and pipeline comparison 2-11 versus TI SDRAM 2-10 R L latency 2-4, 2-7, 3-4, 3-13, 3-17, 3-26, 3-38, 3-48 M MASK command 2-7, 3-26, 3-48 masking 2-7, 3-23, 3-48 mode-register set 2-8 MRS command 2-8, 3-2 N nCCD specification 2-10, 3-21, 3-23, 3-26, 3-28, 3-30, 3-32, 3-34, 3-36 nCLE specification 3-38, 3-39 nDOD specification 3-23, 3-48 nEP specification 3-5 to 3-8, 3-9 to 3-12, 3-28 nHZP specification 3-28 NOOP command 2-6, 3-2, 3-44 nRSA specification 3-2 nWCD specification 3-13 overview of the SDRAM read with bank interleaving -- random row address autodeactivate 3-9 to 3-10 with bank interleaving -- seamless 3-4 to 3-11 read burst interrupted by a DEAC / DCAB command 3-28 interrupted by a READ command 3-21 interrupted by a STOP command 3-26 interrupted by a WRT command 3-23 to 3-25 READ command 3-4 to 3-8, 3-21, 3-23, 3-26 to 3-28, 3-28 to 3-31 read latency 1-3, 3-4 READ-P command 2-6, 3-9 to 3-12 interruption not allowed 3-21 reduced-power operating mode 3-42 See also power-down mode REFR command 2-9, 3-2, 3-46 refresh 2-9 CAS-before-RAS (CBR) refresh 2-9 self refresh 2-9 to 2-12 row-address entry. See bank activation S O 2-2 to 2-3 P PDE command 2-7, 3-42, 3-44 ping-pong operation 2-5 power-down mode (PDE command) 2-7, 3-38 power-down-mode entry -- both banks active 3-42 power-down-mode entry -- both banks deactivated 3-44 to 3-46 power-up sequence 3-2 to 3-4 Index-2 precharge (bank deactivation), 1-2, 2-6 program features burst sequence 2-3 burst type 2-3 SDRAM block diagram 2-2 overview 2-2 versus DRAM 1-2 seamless (gapless) operation 2-5, 3-4, 3-5, 3-21 seamless read with bank interleaving 3-4 to 3-6 self-refresh operation 2-9, 3-46 SLFR command 2-9, 3-46 STOP command 3-26, 3-34 penalty associated with 3-26 synchronous DRAM definition 1-2 system clock input (CLK) 2-2 Index T throughput 1-3 two-bank column-access operation 2-5 two-bank operation 2-5 two-bank row-access operation 2-5 W write burst interrupted by a DEAC / DCAB command 3-36 to 3-37 interrupted by a READ command 3-30 to 3-31 interrupted by a STOP command 3-34 to 3-35 interrupted by a WRT command 3-32 WRT command 2-5, 3-13, 3-23, 3-26, 3-32, 3-36 WRT-P command 2-6, 3-17, 3-23, 3-30, 3-32 interruption not allowed 3-30 write with bank interleaving 3-13 to 3-14 with bank interleaving -- random row address autodeactivate 3-17 to 3-18 Index-3 Index-4