A seasoned veteran in the semiconductor industry, Desi Rhoden speaks at VIA Technology Forum this week in Taipei, and interview was conducted with him to discuss current and future developments of memory technologies.

Q: The DDR2 standard was published by JEDEC in September 2003, but the mass adoption of the memory is only expected in the second half of this year. Does this follow your initial expectations? Do you think two years is the normal timeframe between standard ratification and its adoption by the industry?

A: Two years from specification to widespread usage is normal. System development takes time and even though a lot of work goes on in parallel, most of the earliest systems that utilize a new technology such as DDR2 do not take advantage of all the features that make that newer technology appealing. Often second spins on controllers and systems increase the speed and feature usage, thus making the newer technology more desirable to the end user.

Q: What is the current status of the DDR2-800 specification? When do you expect this specification to be finally ratified by JEDEC?

A: DDR2-800 is on its way and will become widespread, especially with the shift in DRAM technology to 90nm. The timeframe is mostly driven by fab timing and the availability of controllers and systems to support DDR2-800, which will possibly arrive at around the end of 2005 or beginning of 2006.

Q: What do you see as the next stage – DDDR2-933? DDR2-1066?

A: The next speed grade will likely be 1066MHz and because of internal limitations that performance range will probably be limited to the overclocking and gaming markets.

Q: Standard JEDEC timings are usually slower than what DRAM-module makers actually later achieve with their products. Do you think this is normal? Why does JEDEC not ratify faster timings in its standards?

A: The workload to standardize any speed grade is very large for JEDEC, and the focus is almost always confined to the most mainstream speeds, like 667MHz and 800MHz, for DDR2. JEDEC rarely spends time on the fringes like 1066MHz. It is normal for the module manufacturers to push the limits, especially those interested in the gaming markets that have smaller numbers, but are willing to pay a premium for the performance.

Q: Has JEDEC already started DDR3 standard development? What is the current status?

A: JEDEC is already well along in the development of the DDR3 standard, and we have been working on it for about three years now. Initial investigations have already started on memory technology beyond DDR3. JEDEC always has about three generations of memory in various stages of the standardization process: current generation, next generation, and future.

Q: What are the main technology improvements in DDR3? How does it work?

A: Virtually all mainstream standard memory technologies are just incremental improvements over their previous versions. DDR3 is no exception. It is very much like DDR2 except it has lower power, and higher performance. It is always reasonable to expect that the next generation of any standard memory will ultimately gain about two times in raw performance and usually a few enhancements to make the actual performance even better than that. Expect the same from DDR3 all the way up to DDR3-1600. Reduced power consumption has also been a major focus as well in the development of DDR3.

Q: Can you already reveal any details about DDR3, such as bus width, operating voltage and number of pins per module? When do you expect DDR3 to be adopted by the mass market?

A: For the questions you ask, DDR3 looks very much like DDR2 in that it has x4, x8 and x16 organizations and supports almost the same functions. The supply voltage will be 1.5V, and the modules, including unbuffered DIMM, registered DIMM and FB-DIMM2, will be physically the same size and shape, but with a few more pins for functionality. Following historical models, you could reasonably expect the same three-year transition to a new technology that you have seen for the last several generations of standard memory. Process improvements, controllers and systems, all have to be developed in parallel with the devices themselves, and the specification is not yet finished.

Q: You have mentioned FB-DIMM2. What can you say about this technology now?

A: In order to support DDR3 it will require a new AMB device with some new and, as yet, undefined features including a higher-frequency interface.

Q: What do you see after DDR3? Another DDR-based technology? Another kind of technology? Serial or parallel?

A: It is a little early to speculate about the next generation when we have not yet completed DDR3. Many ideas are under discussion, and I am sure even more will be discussed before any decisions are made.

Q: Do you think the idea of logic-combined DRAM has finally been rejected by the industry and will never be taken up again?

A: I have learned that “never” is a dangerous word and I try to avoid it. The idea has been around for a very long time and I expect at least a few out there to continue to work with it looking for a good home. Clearly the majority of the market seems to prefer the stand alone approach, but one should not forget that DRAM today has a great deal of control logic associated with the interface already on the same die.

Q: Do you think manufacturing-process limitations may hinder DRAM-technology progress? Do you see any solutions to speed up the progress in a similar way to the multi-core processors?

A: Process technology always seems to find solutions when the industry needs them. I do not see any real limitations for the foreseeable future. DRAM had multi-bank long before processors had multi-core.

Q: What can you say about claims from Rambus that their patented technologies, which were available for use by JEDEC members when Rambus was a member of the body, were being illegally used once Rambus left JEDEC? Are the DDR2 and DDR3 standards free from this sort of claims by Rambus?

A: I really have nothing to say. These issues are being worked out in the courts.

Q: Do you think there is any chance for Rambus and JEDEC to cooperate with each other again?

A: Anything is possible.

Q: What do you think about Intel’s role in the memory-standard developments? I heard Intel said they would prefer to see JEDEC processes be slower and more mature. Can you comment on this?

A: Intel is one of many companies actively involved in the development of memory standards. I have heard no requests for JEDEC to move slower. The processes in JEDEC are continuously improving and are quite mature having started in 1924 with the Radio Manufacturers Association. JEDEC has been standardizing memory since 1974, one year after the invention of the single transistor memory cell. Today the entire process is completely automated and without paper, instantly available online around the world anytime of day or night. Maybe you heard that the JEDEC process is much faster than the speed at which engineers work, which is true. The bottleneck today is the engineering time required to evaluate and develop material.

Q: What is your opinion about Intel’s validation program for DRAMs and DRAM modules? Does it help or hinder JEDEC’s work?

Several companies have their own validation programs and, like Intel, most are originally based on JEDEC standards for devices and modules. The programs normally have no impact on JEDEC unless one of them uncovers an issue with a JEDEC specification. In that case, the JEDEC companies work together to resolve the issue and modify or clarify the specification if necessary.

Q: Will JEDEC standardize FB-DIMM technology including parts such as AMB? Do you or plan to cooperate with Memory Implementers Forum?

A: JEDEC has been working on the development of FB-DIMM and AMB standards for a long time, and that work is continuing at this time.

Q: Do you agree that soon or later FB-DIMM will be dominating the server-memory technology? What is your opinion about FB-DIMM prospects in the desktop market?

A: I try not to speculate about dominant technologies and prefer to let the users decide as they always do anyway. DDR2-667 and DDR2-800 registered DIMMs will most likely be quite desirable by the server market because they are such an easy choice, since they follow today’s existing model and JEDEC has completed most of the work there already. FB-DIMM will also be available in almost the same timeframe, so in the end the users will decide. For DDR2 desktop 667MHz and 800MHz unbuffered DIMMs, systems are already planned and follow the same existing platform configurations in use today, and there is not much room for FB-DIMM. In the end, all of us in the supplier community do not really care what the users pick, since the market is all about selling more memory modules regardless of their configuration.

Q: What can you say about Taiwan-based companies, DRAM vendors and DRAM-module makers, working with JEDEC? Would you like to see them more active in JEDEC developments?

A: Many Taiwan companies are already actively involved in JEDEC. Committee meetings have even been held in Taipei in the past. There is always room for more companies and more is better. Many times, I have witnessed the benefit of having many companies developing a standard and each one making the end result better.

Q: Some experts expect this year to be tough for both DRAM vendors and DRAM-module makers. Do you agree? Do you have any expectations for the future years?

A: DRAM is a man made commodity and the market is almost entirely driven by supply demand forces in the industry. It is not a business for anyone with a weak heart. Those of us that are in the industry thrive on the innovation and challenge and we have learned that the market continually changes. A year is a long time and there will no doubt be some good and some bad as there always is. The transition to DDR2 should bring some good results for DRAM companies toward the end of this year and the beginning of next. Next year should be especially good for DDR2.

Q: What is your forecast about memory type and size for typical notebook, desktop and server products at the end of this year and the end of 2006?

A: The transition from DDR to DDR2 will continue and accelerate through next year in all segments. The baseline for systems is quickly becoming 512MB.

Desi Rhoden is a guest speaker at VIA Technology Forum (VTF), June 1-3 in Grand Hyatt Hotel Taipei. More information about the VTF is available through www.via.com.tw.

A seasoned veteran in the semiconductor industry, Desi Rhoden now serves as executive vice president at Inphi, a privately-held fabless electronic-components company located in the US and focused on high-performance analog devices such as registers, phase lock loops (PLLs) and advanced memory buffers (AMBs). He is also chairman of the JC-42 Memory Committee and the past chairman of the board for JEDEC Solid State Technology Association, the semiconductor engineering standardization body of the Electronic Industries Alliance (EIA). Prior to Mr. Roden’s speech at VIA Technology Forum this week in Taipei, DigiTimes.com spoke to him to discuss current and future developments of memory technologies.

Q: The DDR2 standard was published by JEDEC in September 2003, but the mass adoption of the memory is only expected in the second half of this year. Does this follow your initial expectations? Do you think two years is the normal timeframe between standard ratification and its adoption by the industry?

A: Two years from specification to widespread usage is normal. System development takes time and even though a lot of work goes on in parallel, most of the earliest systems that utilize a new technology such as DDR2 do not take advantage of all the features that make that newer technology appealing. Often second spins on controllers and systems increase the speed and feature usage, thus making the newer technology more desirable to the end user.

Q: What is the current status of the DDR2-800 specification? When do you expect this specification to be finally ratified by JEDEC?

A: DDR2-800 is on its way and will become widespread, especially with the shift in DRAM technology to 90nm. The timeframe is mostly driven by fab timing and the availability of controllers and systems to support DDR2-800, which will possibly arrive at around the end of 2005 or beginning of 2006.

Q: What do you see as the next stage – DDDR2-933? DDR2-1066?

A: The next speed grade will likely be 1066MHz and because of internal limitations that performance range will probably be limited to the overclocking and gaming markets.

Q: Standard JEDEC timings are usually slower than what DRAM-module makers actually later achieve with their products. Do you think this is normal? Why does JEDEC not ratify faster timings in its standards?

A: The workload to standardize any speed grade is very large for JEDEC, and the focus is almost always confined to the most mainstream speeds, like 667MHz and 800MHz, for DDR2. JEDEC rarely spends time on the fringes like 1066MHz. It is normal for the module manufacturers to push the limits, especially those interested in the gaming markets that have smaller numbers, but are willing to pay a premium for the performance.

Q: Has JEDEC already started DDR3 standard development? What is the current status?

A: JEDEC is already well along in the development of the DDR3 standard, and we have been working on it for about three years now. Initial investigations have already started on memory technology beyond DDR3. JEDEC always has about three generations of memory in various stages of the standardization process: current generation, next generation, and future.

Q: What are the main technology improvements in DDR3? How does it work?

A: Virtually all mainstream standard memory technologies are just incremental improvements over their previous versions. DDR3 is no exception. It is very much like DDR2 except it has lower power, and higher performance. It is always reasonable to expect that the next generation of any standard memory will ultimately gain about two times in raw performance and usually a few enhancements to make the actual performance even better than that. Expect the same from DDR3 all the way up to DDR3-1600. Reduced power consumption has also been a major focus as well in the development of DDR3.

Q: Can you already reveal any details about DDR3, such as bus width, operating voltage and number of pins per module? When do you expect DDR3 to be adopted by the mass market?

A: For the questions you ask, DDR3 looks very much like DDR2 in that it has x4, x8 and x16 organizations and supports almost the same functions. The supply voltage will be 1.5V, and the modules, including unbuffered DIMM, registered DIMM and FB-DIMM2, will be physically the same size and shape, but with a few more pins for functionality. Following historical models, you could reasonably expect the same three-year transition to a new technology that you have seen for the last several generations of standard memory. Process improvements, controllers and systems, all have to be developed in parallel with the devices themselves, and the specification is not yet finished.

Q: You have mentioned FB-DIMM2. What can you say about this technology now?

A: In order to support DDR3 it will require a new AMB device with some new and, as yet, undefined features including a higher-frequency interface.

Q: What do you see after DDR3? Another DDR-based technology? Another kind of technology? Serial or parallel?

A: It is a little early to speculate about the next generation when we have not yet completed DDR3. Many ideas are under discussion, and I am sure even more will be discussed before any decisions are made.

Q: Do you think the idea of logic-combined DRAM has finally been rejected by the industry and will never be taken up again?

A: I have learned that “never” is a dangerous word and I try to avoid it. The idea has been around for a very long time and I expect at least a few out there to continue to work with it looking for a good home. Clearly the majority of the market seems to prefer the stand alone approach, but one should not forget that DRAM today has a great deal of control logic associated with the interface already on the same die.

Q: Do you think manufacturing-process limitations may hinder DRAM-technology progress? Do you see any solutions to speed up the progress in a similar way to the multi-core processors?

A: Process technology always seems to find solutions when the industry needs them. I do not see any real limitations for the foreseeable future. DRAM had multi-bank long before processors had multi-core.

Q: What can you say about claims from Rambus that their patented technologies, which were available for use by JEDEC members when Rambus was a member of the body, were being illegally used once Rambus left JEDEC? Are the DDR2 and DDR3 standards free from this sort of claims by Rambus?

A: I really have nothing to say. These issues are being worked out in the courts.

Q: Do you think there is any chance for Rambus and JEDEC to cooperate with each other again?

A: Anything is possible.

Q: What do you think about Intel’s role in the memory-standard developments? I heard Intel said they would prefer to see JEDEC processes be slower and more mature. Can you comment on this?

A: Intel is one of many companies actively involved in the development of memory standards. I have heard no requests for JEDEC to move slower. The processes in JEDEC are continuously improving and are quite mature having started in 1924 with the Radio Manufacturers Association. JEDEC has been standardizing memory since 1974, one year after the invention of the single transistor memory cell. Today the entire process is completely automated and without paper, instantly available online around the world anytime of day or night. Maybe you heard that the JEDEC process is much faster than the speed at which engineers work, which is true. The bottleneck today is the engineering time required to evaluate and develop material.

Q: What is your opinion about Intel’s validation program for DRAMs and DRAM modules? Does it help or hinder JEDEC’s work?

Several companies have their own validation programs and, like Intel, most are originally based on JEDEC standards for devices and modules. The programs normally have no impact on JEDEC unless one of them uncovers an issue with a JEDEC specification. In that case, the JEDEC companies work together to resolve the issue and modify or clarify the specification if necessary.

Q: Will JEDEC standardize FB-DIMM technology including parts such as AMB? Do you or plan to cooperate with Memory Implementers Forum?

A: JEDEC has been working on the development of FB-DIMM and AMB standards for a long time, and that work is continuing at this time.

Q: Do you agree that soon or later FB-DIMM will be dominating the server-memory technology? What is your opinion about FB-DIMM prospects in the desktop market?

A: I try not to speculate about dominant technologies and prefer to let the users decide as they always do anyway. DDR2-667 and DDR2-800 registered DIMMs will most likely be quite desirable by the server market because they are such an easy choice, since they follow today’s existing model and JEDEC has completed most of the work there already. FB-DIMM will also be available in almost the same timeframe, so in the end the users will decide. For DDR2 desktop 667MHz and 800MHz unbuffered DIMMs, systems are already planned and follow the same existing platform configurations in use today, and there is not much room for FB-DIMM. In the end, all of us in the supplier community do not really care what the users pick, since the market is all about selling more memory modules regardless of their configuration.

Q: What can you say about Taiwan-based companies, DRAM vendors and DRAM-module makers, working with JEDEC? Would you like to see them more active in JEDEC developments?

A: Many Taiwan companies are already actively involved in JEDEC. Committee meetings have even been held in Taipei in the past. There is always room for more companies and more is better. Many times, I have witnessed the benefit of having many companies developing a standard and each one making the end result better.

Q: Some experts expect this year to be tough for both DRAM vendors and DRAM-module makers. Do you agree? Do you have any expectations for the future years?

A: DRAM is a man made commodity and the market is almost entirely driven by supply demand forces in the industry. It is not a business for anyone with a weak heart. Those of us that are in the industry thrive on the innovation and challenge and we have learned that the market continually changes. A year is a long time and there will no doubt be some good and some bad as there always is. The transition to DDR2 should bring some good results for DRAM companies toward the end of this year and the beginning of next. Next year should be especially good for DDR2.

Q: What is your forecast about memory type and size for typical notebook, desktop and server products at the end of this year and the end of 2006?

A: The transition from DDR to DDR2 will continue and accelerate through next year in all segments. The baseline for systems is quickly becoming 512MB.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved