Wednesday, May 22, 2002 10:14:59 PM
Memories of the future
Maury Wright, Editor-in-Chief -- 5/1/2002
CommVerge
At the edge of a converged network, you'll find a variety of intelligent devices that span applications from entertainment to productivity and reside everywhere from the auto to the living room. All of these nodes share some characteristics, such as connectivity and intelligence, and they all rely on some of the same key enabling technologies.
Nonvolatile memory is perhaps the most important of the enablers, even if processors get the most glory. Flash memory serves in cell phones, set-top boxes, digital music players, and a host of other devices, acting both as program storage and as a content store for music, pictures, contact lists, and many other data types.
Given the significance of flash, we decided to host a roundtable discussion on the topic. We felt that such a format might prove valuable because it would allow industry experts to pontificate on the issues directly. The summit took place only virtually—via email—but yielded a robust, realistic dialogue nonetheless.
Follow along to learn where flash-memory capacity and prices are headed, which applications will drive consumption, whether alternative nonvolatile memories will encroach upon flash markets, and other valuable insights.
CV: Because CommVerge generally focuses on convergence applications and uses that application-level focus to spotlight enabling technologies like memory, I'd like to start at the application level. Could each of you describe the three or four products that consume the largest quantities of flash memory today?
PARTICIPANTS
Philippe Berge, Director of Marketing, STMicroelectronics Memory Product Group
Bertrand Cambou, Group Vice President, Memory Group, Advanced Micro Devices
Keith Horn, Vice president of Marketing, Fujitsu Microelectronics
Bill Krenik, Wireless Advanced Architecture Manager, Texas Instruments
Brian Kumagai, Business Development Manager, Flash Products, Toshiba
Kevin Plouse, Vice President of Technical Marketing and Business Development, Memory Group, Advanced Micro Devices
Sudeep Sharma, Associate Vice President, Memory Division, Mitsubishi Electric and Electronics USA
Victor Tsai, Product Marketing Manager, Flash Products, Hitachi
Mike Williams, Director of Marketing, Flash Products Group, Intel
Bing Yeh, President and CEO, Silicon Storage Technology
Sudeep Sharma (Mitsubishi): Today the largest quantities of flash memory are consumed in cellular handsets, storage cards, BIOS flash applications for PCs, and portable electronic devices such as digital cameras, PDAs, and MP3 players.
Kevin Plouse (AMD): Cellular telephones use the bulk of flash memory today, and we don't expect that to change anytime soon. So, when we look in our crystal ball, we see a cell phone with more features that uses more flash memory. One key point is that the people who invested in 3G networks invested that money because those networks drive data, and they drive data to phones. The convergence of the phone and the handheld computer is the single largest opportunity for flash memory. The second driver is the consumer product. There are all kinds, but the ones that come to mind are music, video, and photo storage (cameras, video recorders, etc). If the cost is right, these will drive a lot of demand. Consumer appliances too—like DVD players, high-definition television—drive growing demand for flash devices.
Bertrand Cambou (AMD): And the third one is internetworking. Obviously, the dot-com explosion resulted in extensive investments in networks, and flash provided network reprogrammability. The dot-com explosion has been replaced with a dot-com collapse, but eventually the networks will be replaced. We don't know yet what they will look like, but we can assume that with the fundamental growth in data, we must keep data moving.
Mike Williams (Intel): Cellular phones by far consume the most flash in millions of megabytes. Cell phones are the highest volume, shipping approximately 400 million units with a large density mix. Digital cameras would be next, not because of their volume, but due to their higher average density. The next few applications—which include networking/communications, set-top boxes and handhelds—are all smaller and comparable in consumption.
CV: Today, the flash market is clearly divided into data- and code-storage segments, dominated by NAND/AND [not and/and] and NOR [not or] flash architectures, respectively. How do these different architectures match up with the flash applications? Also, please explain the need for flash in these applications and point out the key memory-system requirements, such as capacity, speed, cost, and others.
Brian Kumagai (Toshiba): Today's primary NAND applications are digital cameras (mainly in removable-card format), game consoles/accessories, and digital audio. Today's primary NOR applications are cell phones, PDAs, and set-top boxes. NOR applications are increasingly being restricted to code storage/execution, where the density requirements are relatively small, code must be executed from flash, and write performance/reliability are not concerns.
“The people who invested in 3G networks invested that money because those networks drive data, and they drive data to phones. The convergence of the phone and the handheld computer is the single largest opportunity for flash memory.”
Kevin Plouse,
Advanced Micro Devices
Mike Williams (Intel): Flash is critical to all these applications for supplying system and application code and data storage. The best way to split the flash market requirements is between code and code+data architectures/requirements (cellular, handheld, set-top, networking, and telematics) and pure data-only storage (digital cameras and MP3 players). Code and data applications typically require high-performance reads (burst or page-mode), read-while-write capability at 66-MHz, data integrity/reliability, low power, and mid- to high-density capacity, while data-only applications require and value high density in a removable form factor.
Philippe Berge (STMicroelectronics): In addition to mobile terminals (cell phones), we see PC BIOS, automotive, and digital home gateways as key markets. In mobile terminals, the key requirements are low power consumption, high-density, tiny packages and footprint, an optimized interface with the baseband processor, and the ability to combine flash with SRAM. These nonvolatile memory requirements are directly driven by more and more user-friendly application features, such as GPRS [general packet radio services] and WAP [wireless application protocol], Internet and talk-mode protocols, tri-band support, voice memos, voice recognition, predictive text input, and color displays.
Games require bigger and bigger operating systems, hence bigger and bigger nonvolatile memory that has to be executed as fast as possible. Longer standby and talk times require low-power supply and low-power operative and standby consumption. In digital home gateways, the key requirements are cost and write and programming throughput. They are driven by the following system evolution: Web navigation, e-commerce, expert systems for user profiling, and remote software downloads for things like operating-system updates and TV program guides.
CV: From a pure silicon perspective, discuss your organization's technology roadmap in the NAND and/or NOR camps. Tell me where you stand today in terms of capacity and where you expect to be in 2005. Please describe, at a high level, the techniques that will deliver on your roadmap, such as multilevel cells (MLCs) as opposed to single-level cells (SLCs).
"Until a product is proven in production with real customers, it is difficult to place much faith in it. MRAM has been researched for over 30 years, yet it is still not in mass production."
Bing Yeh, Silicon Storage Technology
Bing Yeh (Silicon Storage Technology): SST is aligned in the NOR camp, and we believe this will continue to be the dominant area for flash, especially for code storage, but also for data storage. Code storage requires fast access times for system boot-up and reliable byte access without the latency that is common in NAND flash. Furthermore, in low density, NAND cannot compete, as it requires massive overhead circuitry to implement.
MLC technology will bridge the gap in cost between NAND and NOR in the medium and higher densities, and we foresee a realistic roadmap to four bits per cell using MLC SuperFlash technology. Currently, SST has a wide range of capacity in the low densities from 256 kbits to 16 Mbits. We will expand into the medium densities, from 32 Mbits and up, for the coming years for the code-storage applications. We also plan to offer more than 1 Gbit per chip for the mass data-storage applications.
Sudeep Sharma (Mitsubishi): We are primarily focused on DINOR technology, a special type of NOR architecture. Relative to NOR flash, our DINOR technology offers faster random access at a lower voltage, and seven to 25 times quicker erase cycle. All of our flash-memory parts also have a BGO (background operation) function. Mitsubishi was the first to adopt the BGO function in 1997 on 8-Mbit flash. BGO can eliminate EEPROM [electrically erasable programmable read-only memory] from cellular phones since data can be read from banks while another bank is being programmed or erased.
Mike Williams (Intel): Our product portfolio is focused on NOR, not only for code but also for specifically optimized code+data requirements. We have three product lines. Our high-performance Wireless Flash, for handheld customers requiring the ultimate performance, offers a 1.8-volt (3-volt I/O option) product family with densities from 32 to 128 Mbits. Currently in production on 0.18-micron technology, we are sampling now on 0.13-micron, with a roadmap to 90 nanometers. Also on 0.13-micron, we are adding a new x32 implementation and increasing density to 512 Mbits by 2005.
Intel StrataFlash Memory is the highest-density, lowest-cost flash memory for code+data applications. Used in nearly every WinCE/PocketPC handheld, today's StrataFlash memory on 0.18-micron is Intel's third generation of MLC technology, which we originally introduced in 1997. StrataFlash is offered today in 32- to 128-Mbit densities and a 256-Mbit density later this year at 3 volts (1.8-volt I/O available). A high-performance 1.8-volt version will be released later this year, and densities on the MLC technology will reach 512 Mbits by 2005.
“The anticipated largest consumers by 2005 should be cell phones, consumer electronics (digital cameras, MP3 players), networking, and automotive (including engine control and navigation systems).”
Keith Horn, Fujitsu
Industry-standard boot block (C3/B3) flash, now in its fourth generation of complete backward compatibility, is currently in production on a 0.13-micron process. This product family includes 8 to 64 Mbits, and production will continue through 2005 and beyond. In addition to continuous improvement, leading lithographies that keep us one product generation ahead of our nearest competitors, and proven multilevel cell manufacturing, we are exploring the use of four bits per cell and Ovonyx Unified Memory to expand our roadmap in the coming years.
Brian Kumagai (Toshiba): In NAND flash, we are currently in mass production of 512-Mbit monolithic SLC, 1-Gbit stacked (two-chip) SLC, and 1-Gbit monolithic MLC. In 2005, maximum density will increase to 4-Gbit SLC and 8-Gbit MLC monolithic devices. In NOR flash, our highest density today is a 128-Mbit SLC. We have plans for 256-Mbit and possibly 512-Mbit MLC devices in 2005.
Kevin Plouse (AMD): FASL [Fujitsu AMD Semiconductor Limited, AMD's joint venture with Fujitsu] is a leader in NOR technology. We are neck-and-neck between FASL and Intel for first and second position in the market. The NAND/NOR line is getting more and more blurry in terms of applications. Our customers prefer NOR but they want cost reductions.
Bertrand Cambou (AMD): One path to cost reduction is MLC.... [However,] we don't see how 4-bit MLCs can work reliably for code-storage solutions. As a result, we see the classical floating-gate technology coming to a point where it is not extendable anymore. That is why AMD took a different path with our MirrorBit architecture, which is not based on the MLC principle.... For years we have worked to develop an alternative path and now we are working full speed on MirrorBit—a technology without the compromises associated with MLC. We also recognize that MirrorBit is very expandable, even to four bits per cell. We believe that the move away from floating-gate will happen and our conviction is strong that we are engaged in a paradigm shift.
“There will be a sustainable need for code-storage flash that will be driven by the need for bigger and bigger operating systems enabling more and more user-friendly applications.”
Philippe Berge,
STMicroelectronics
Victor Tsai (Hitachi): We are a major supplier of data-storage flash with our MLC AND-type flash technology, and we are a manufacturer of code-storage NOR flash products. Hitachi recently introduced the new AG-AND multilevel flash memory cell, which gives Hitachi a technology and cost advantage over competing data-storage flash products and technologies.
Keith Horn (Fujitsu): We currently offer only NOR flash. However, our Multi Chip Package lineup will continue to provide both NOR and NAND flash. The company's flash roadmap offers an impressive range of densities (2 to 128 Mbits), voltages (5 to 1.8 volts), and we have a well-established reputation for advanced packaging methods.
CV: Is there the possibility that the flash industry might consolidate toward a single type of flash architecture? For example, could NAND flash be augmented with DRAM cache and control circuits that would allow code-storage applications to leverage the low-cost, high-density benefits of NAND flash? Or, are there breakthroughs in the NOR world that can ramp capacities and lower costs to compete with NAND flash?
Keith Horn (Fujitsu): The disadvantage of NAND flash is its reliability. Some applications simply cannot risk reliability issues and will be forced to continue to utilize NOR or NOR-like flash. However, the production of multibit cell flash product will allow higher NOR-like reliability with pricing that is more in line with NAND flash.
Bing Yeh (Silicon Storage Technology): NAND- and NOR-type applications and specs are quite different, and both types will coexist forever. Four-bit-per-cell MLC will provide a great challenge to NAND flash in terms of cost. However, because several large Japanese companies have focused on NAND flash, there will continue to be some NAND flash market inertia. So, regardless of what arguments technologists might make about whether NOR or NAND is technologically better, NAND will continue to play a role in the high-density flash market. In the embedded and mainstream code-storage markets, however, NAND will never penetrate. We see clear evidence of this, as NAND vendors have followed a DRAM model in the manufacture of NAND flash, pushing products into higher and higher density and not even offering NAND devices anywhere below 64 Mbits, which is the domain of code storage today.
Victor Tsai (Hitachi): There may be a point in time where there would be a convergence of data-storage and code-storage flash. Data-storage flash is generally more cost-effective than code-storage flash. Hitachi has just introduced the superAND flash product, which incorporates some NOR-like features, including power-on read for system boot-up and 100 percent good memory without error handling and memory management by the host CPU. This is the first crossover product that can satisfy both data-storage and code-storage needs in a system.
Sudeep Sharma (Mitsubishi): We don't see the NOR and NAND types of flash-memory architectures converging. However, new flash-memory architectures may be developed to handle both types of applications.
Mike Williams (Intel): We believe it will continue to fragment. Application requirements are diverging rather than converging. We see this today in multiple line-item offerings on our silicon and numerous stacking configurations requested. Additionally, our long-term strategic alignment with our top customers indicates continued diversification. One size certainly does not fit all.
"Application requirements are diverging rather than converging. Our long-term strategic alignment with our top customers indicates continued diversification. One size certainly does not fit all.”
Mike Williams, Intel
And let me correct a potential misconception with our Intel StrataFlash memory on leading-edge lithographies. We believe we do compete with NAND on a cost basis. The question isn't about cost per se, but about what price a manufacturer is willing to sell that flash device. Currently, NAND manufacturers are pursuing a very aggressive pricing strategy to make up for what we believe is an inherent mismatch with the system requirements in a code or code+data environment (bad blocks, error correction, read speeds, increased system memory, etc).
CV: Mike, is it your point that NAND manufacturers have cut prices to artificially low levels to gain entry into code or code+data storage applications, and that some buyers will deal with mismatched characteristics like slow read speeds to buy the lower-cost flash? And when you say you "compete with NAND on a cost basis," are you making that claim based on system costs in a code or code+data application?
Mike Williams (Intel): There were some lofty expectations for NAND growth the past few years, mainly driven by growth projections for digital cameras and digital music players. Each year, the forecasts were pushed out another year. The missed growth expectations have left the NAND suppliers scrambling to find homes for their products, and they have resorted to trying to fit their products into the traditional NOR markets. But NAND has inherent feature mismatches for these applications. For example, you cannot execute out of NAND, given the slow read speeds. Therefore, redundant memory, consuming more space and power, is required for the device to operate. NAND also requires error-correction circuitry. NAND contains bad blocks that must be managed. And the list goes on.
"In the near-term, the 'perfect' memory, nonvolatile RAM, will remain an R&D product. While some technologies appear promising, we believe the applications will be restricted."
Brian Kumagai, Toshiba
In short, there are a number of system-complexity issues when designing with NAND, and the NAND suppliers are attempting to overcome these issues by using cost as an incentive. Hence, NAND is selling at a very aggressive price today. In most cases involving both code and code+data applications, Intel StrataFlash memory offers a lower overall system cost and is much easier to use in design.
CV: Given the state of the market today, and developments in NAND and NOR technologies, take a look at your crystal ball and project the top three or four products for 2005. And again, please explain the key memory-system requirements for flash.
Keith Horn (Fujitsu): The anticipated largest consumers by 2005 should be cell phones, consumer electronics (digital cameras, MP3 players), networking, automotive (including engine control and navigation systems). As cell phones offer more and more features, they will continue to require larger densities of memories and smaller packages.
Mike Williams (Intel): By 2005, cellular, cameras, networking, PDAs, and set-top boxes will remain as the top markets, in our opinion, with cellular continuing to lead and handheld growth most likely outpacing the others. Telematics/GPS will also emerge as a top flash application.
Brian Kumagai (Toshiba): In 2005, NAND applications will include digital still/video cameras, cell phones (mainly for digital camera/audio/video purposes), PDAs, and set-top boxes. In all of these applications, whether the flash is used for code and/or data storage, the primary factors driving the usage of NAND are the requirements for high density and low cost. Additionally, for the data-intensive applications, the superior write performance and reliability of NAND compared with NOR is an important consideration. NOR applications for 2005 include cell phones, low-end set-top boxes, and networking/communications equipment.
Kevin Plouse (AMD): Looking into our crystal ball though, we can't forget to talk about the auto dashboard. It's small, but the fastest growing forecast is in the cockpit of the car—for entertainment and navigation. The car PC has just hit the inflection point for growth. It's been in development for eight years or more and is now becoming a standard part of the car.
“The car is a very interesting environment for us because we have been focusing on the car for a while, and you do not use substandard flash in a car.”
Bertrand Cambou,
Advanced Micro Devices
Bertrand Cambou (AMD): The car is a very interesting environment for us because we have been focusing on the car for a while, and you do not use substandard flash in a car (for example, because of the extreme temperature variation requirements). And that has been our strength.
CV: With cell phones identified as such a huge consumer of flash memory, could you further illuminate how flash is used in those devices. Digital cell phones rely on high-speed DSPs, so I know SRAM is required for code execution. Perhaps you could provide a scenario for what types of code and data are stored in different memory types, both when a phone is standing by and when a call is in progress. And describe how close this model of memory usage comes to other applications like PDAs or telematics systems.
Mike Williams (Intel): Flash memory has traditionally been used in cellular handsets to store program code used to control the operation of the device, to store data for device-tuning parameters, and to store data such as frequently used phone numbers and other personal information. Flash was adopted in these devices due to its solid-state ruggedness and high data retention—a phone can be dropped to the floor, the battery can be removed, and the information in the flash memory is retained.
Internet capable handsets, including new 2.5G and 3G phones, are driving the requirements for higher-performance and higher-density flash memory. These cellular handsets can be separated into two main processing functions: the baseband communications processor and the applications processor. Flash memory is used in the baseband unit to store program code for the traditional microcontroller device in charge of handling the specific cellular protocol. Flash memory can also be used in the baseband unit to store the DSP algorithms, as well as acting as the main memory in the event of an onboard cache miss from the integrated SRAM memory. Regardless of standby operation or active operation, the baseband processor is continually executing code from the flash device. In standby mode, approximately 1 to 3 percent of the time (depending on the actual protocol), the baseband processor must "wake up" to ping the nearest basestation in order to stay connected.
Flash-memory requirements are exploding on the application-processor side, where flash is used to store program code for new functionality such as Web browsers, color displays, Java applets, and audio/digital data manipulation. Connecting to the Internet opens up the need for more data on the application-processor side for storing large video files, digital music files, photographs, and email.
The memory usage between an application processor in a cell phone and a PDA are the same (hence the convergence of cell phones and PDAs). The industry debates whether one common multipurpose device will emerge or whether we will continue to see a variety of devices tailored for a specific need. Whether a cell phone, PDA, or telematics, Intel is offering common building blocks, including baseband processors, applications processors, and flash memory based on the Intel Personal Internet Client Architecture—a development blueprint for wireless devices and software combining voice and data.
CV: Brian Kumagai of Toshiba seems to imply that future cell phones will have a mix of NAND and NOR flash. I assume the former will serve integrated add-on functionality like a digital camera or MP3 player, while the latter, I assume, serves to store code for the cell-phone application. Could you give me a precise picture of how you see this memory architecture applied?
“Although in many ways integration is the key to low cost, we believe that discrete flash memory will continue to be less expensive than embedded flash memory on a per-bit basis.”
Sudeep Sharma, Mitsubishi
Brian Kumagai (Toshiba): We expect both evolutionary and revolutionary cell-phone architectures utilizing NAND flash. The evolutionary architecture will utilize NAND for data (photos, audio, video, etc) and the NOR for all types of code storage. In this case, the NOR will have to be fast enough to support code execution for all processing/control functions, including the DSP, which will probably be realized by page/burst mode. Toshiba would expect the NOR density to increase at about its historical rate for this architecture, since the NAND will take over some of the previous NOR functions, such as phone-number storage. The revolutionary architecture would use only NAND combined with lots of RAM (probably DRAM). In this case, the NAND would store all of the code and data, and the code would be executed out of RAM. The smartphone and PDA-combo phone will drive the transition to the NAND-only architecture.
CV: Outside of external CompactFlash, SmartMedia, SD Card, and Memory Stick modules, will there be a sustainable need for stand-alone flash memory chips going forward? Integration is the key to low cost, and SOC [system-on-chip] is an unmistakable trend. Will flash become largely a feature integrated onto other chips? If not, describe the capacity requirements or silicon limitations that will prevent such consolidation.
Philippe Berge (STMicroelectronics): There will be a sustainable need for code-storage flash that will be driven by the need for bigger and bigger operating systems enabling more and more user-friendly applications. Embedding flash will always remain a tradeoff of cost, footprint, and performance. Overall, flash will keep growing in three directions: standalone, embedded, and cards. Standalone and embedded flash will grow mostly for code storage. Flash cards or flash-plus-other-memory cards will develop as real subsystems for data storage. There is no real standard package yet, but the evolution of a standard associated to cost/bit reduction will push the market to a higher level of volumes/value.
Sudeep Sharma (Mitsubishi): Flash memory is being integrated already and will continue to be more integrated in SOC devices. However, the density of flash memory in SOC applications will continue to be limited because of chip-size constraints, which is also related to the yield issue. Continued development of finer process technologies will increase SOC flash density, but future applications will also continue to demand more flash memory density. Although in many ways integration is the key to low cost, we believe that discrete flash memory will continue to be less expensive than embedded flash memory on a per-bit basis.
Bill Krenik (Texas Instruments): Texas Instruments doesn't make flash, but as a leading vendor of ICs for cellular handsets we have a vested interest in flash developments. For wireless, integrating flash memory on the same chip may result in an inflexible memory configuration, because the handset designer will need to specify the amount of flash memory to be integrated early in the design cycle, before actual memory needs are clear. This may result in excess memory, leading to a cost penalty, or insufficient memory, resulting in loss of product features or the need for added external memory.
Further, flash integration normally requires six to eight additional process reticles over a conventional digital CMOS [complementary metal oxide semiconductor] process, significantly increasing manufacturing costs. Since there are no significant performance benefits obtained by integrating flash onto the same chip for wireless, it is difficult to justify flash integration. Other options, however, such as the use of multidie packaging, may be attractive in some cases.
Bing Yeh (Silicon Storage Technology): SST is by far the world's leading, if not the only substantial vendor, of embedded-flash solutions. Our split-gate SuperFlash architecture is available as integratable intellectual property through many of the world's leading foundries. Today, dozens of blue-chip companies license SST's SuperFlash technology for integration into their own wireless chips and other ASICs on a regular basis. Since SuperFlash technology is CMOS-compatible for fab portability and scalability, and since SuperFlash offers significantly better power usage and die size efficiencies compared with both stacked-gate and NAND flash, SST believes this will continue to be a rapidly growing and successful market.
That being said, however, flash is not always as cost-effectively integrated with other system requirements onto a single chip, due to the additional processing and testing steps required for flash memory. SOC with flash is most effective on very small die, where most of the component cost is in the packaging, or at the very high-density spectrum of so-called smart flash memory, where most of the SOC silicon is occupied by flash.
CV: The mixture of memory technologies on one dedicated memory chip or in a multichip package appears to be another trend in the integration story, and without doubt many convergence products require a mix of memory types. Give me your opinions on what types of mixes will be popular, including the possibility of mixing multiple flash types along with SRAM and DRAM. Describe why a chip dedicated to a mix of memory is a good idea, and if so, how you can craft a standard product family that meets the needs of different applications.
Mike Williams (Intel): Providing one packaged memory subsystem is compelling for handheld devices due to the space savings. Today, we are stacking flash and SRAM into one package, and the possibilities are almost endless for stacking, including flash and flash, flash and logic, flash and other memory, and any of the combinations above. These combinations are driven by the memory-subsystem needs of our customers. Crafting one standard product family is not achievable due to the fragmentation discussed previously. Successful flash suppliers must strive for flexibility and quick turnaround time to meet their customers' specific needs.
Keith Horn (Fujitsu): Our lineup of multichip packaged (MCP) devices, which includes flash and SRAM or flash and Fast-Cycle RAM, will continue to lead the field in mixed-memory technology on one package. For cellular applications, this MCP device can replace multiple components, resulting in space savings. It can also offer higher densities that are not available in today's marketplace with a one-chip solution and at a reasonable cost.
"Rotating storage is not a practical memory solution for today's handsets. However, in the future, the technology may be a good fit for high-end PDAs."
Bill Krenik, Texas Instruments
Bill Krenik (Texas Instruments): In wireless-handset applications, SRAM is normally used for multiple levels of cache, while flash is used for program storage and storage of user data and system settings. Since the cache needs to be integrated with the processor and flash integration appears to be cost prohibitive for wireless, it seems unlikely that SRAM+flash products will emerge.
CV: Do you see any near-term prospects for technologies like MRAM [magnetic RAM], FRAM [ferroelectric RAM], Ovonyx's optical technology, or some other nonvolatile memory to succeed in mainstream applications? Also, can rotating storage technologies, like hard-disk drives and the new Dataplay drive, impact the market for flash modules?
Bing Yeh (Silicon Storage Technology): Until a product is proven in production with real customers, it is difficult to place much faith in it. MRAM has been researched for over 30 years, yet it is still not in mass production.
Brian Kumagai (Toshiba): In the near-term, the "perfect" memory, nonvolatile RAM, will remain an R&D product. While some technologies appear promising, we believe the applications will be restricted. For example, Toshiba is developing 32-Mbit and 64-Mbit FeRAM [FRAM], which can be used to replace NOR+SRAM in low-end cell phones. Still, none of these new technologies will reach the density or cost-per-bit of NAND flash. Toshiba plans to introduce a commercial FRAM by the end of 2002. The target density is 32 Mbits. The primary technical challenge is acceptable performance in terms of access time.
Bill Krenik (Texas Instruments): Of the advanced memory technologies you cite, only FRAM is proven in high-volume manufacturing today. FRAM is also attractive because it can be integrated with very few additional process reticles. While MRAM and Ovonyx memory are very interesting technologies, they remain unproven as real solutions for low-cost, high-volume applications.
Rotating storage, of course, looks great on a cost-per-bit basis. However, this low cost is only available for relatively large memories. As a result, rotating storage is not a practical memory solution for today's handsets. However, in the future, the technology may be a good fit for high-end PDAs.
Mike Williams (Intel): As we've discussed publicly, Intel is pursuing Ovonyx memory technology. Although it is still early in the development, the initial results look encouraging. Compared to MRAM and FRAM, we believe Ovonyx holds the best promise for delivering on the performance, densities, integration, and reliability needed for our customers. If all goes well, we would expect Ovonyx to start making an impact on mainstream applications as early as the middle of the decade. But it is premature to discuss specific product plans.
Rotating storage technology will always be an alternative in the pure data-storage area. We see this in the digital-music-player market segment today, where NAND memory is being squeezed by less-expensive rotating technologies.
CV: We’ll finish with the unpopular question. I’d welcome your views on where flash prices are headed. I’d like to discuss price for two reasons. First, low price enables convergence applications. Music players, for example, have been hampered by flash prices, although I know they’ve dropped considerably (and I know the RIAA [Recording Industry Association of America] has hit the music players harder, but that’s a discussion for another day). However, low price has potentially negative ramifications for flash manufacturers. Moreover, the number of manufacturers making flash today is still relatively large relative to other commodity memory types like DRAM. Is the flash market headed for a major consolidation toward a small group of major players? What characteristics of your business make you a long-term participant in the flash industry?
Bertrand Cambou (AMD): It is our belief and strategy that what we need is to continuously and relentlessly cut the price-per-bit. And to commit to our customers a cost reduction that empowers them to build higher and higher densities into their systems, thereby making flash even more pervasive.
Philippe Berge (STMicroelectronics): We are expecting prices to stabilize in Q2 due to the recovery of the demand and to rise in the second half of the year. As for consolidation, the high end of the code-storage market is already at an advanced stage, with very few suppliers having the proper relationship with the key customers, advanced technology, and manufacturing capacity. High-density flash devices are already coming from only three to four suppliers. Second-tier vendors are shipping devices made with lower-density older technology. In the long term, flash technology is essential for STMicroelectronics’ SOC strategy. Flash offers ST the advantage of both differentiated and standard products. The flash-differentiated products, essentially custom configurations for high-volume applications, are key for our corporate strategic customers and give some stability to the business. The standard-product portfolio contributes by extending our customer base and providing the volume to lower our overall manufacturing costs.
Victor Tsai (Hitachi): There are many code-storage NOR flash suppliers, but the number of data-storage NAND/AND flash suppliers is relatively small. The growth rate of data-storage flash is much higher than code-storage flash, so while there may be consolidation in the code-storage flash supplier base, there is still a lot of market potential for new entries into the data-storage flash market.
Sudeep Sharma (Mitsubishi): We believe flash-memory demand will increase strongly and likely outstrip supply as the US and worldwide economy improves and as cellular handset demand increases. Mitsubishi Electric has been strong for a long time in providing a wide variety of memory technologies that can be combined to provide a complete solution.
Keith Horn (Fujitsu): Flash prices appear now to have stabilized. We have not seen prices increase yet, but they certainly are not decreasing. Low prices may eliminate some newcomers to the flash business, but established flash manufacturers will continue to thrive by implementing die shrinks and investing in new technologies such as multibit cell product. Fujitsu should be considered a long-term participant because of our joint venture with AMD and the considerable investment that has been made in our facilities.
Kevin Plouse (AMD): Flash has attracted every major memory player. Those that are the strongest will survive, the best technology will survive, the most innovative will survive. We’ve been a leader in nonvolatile memory for more than a quarter century. We’ve built a strong partnership with Fujitsu. With Fujitsu, we believe we have the best high-volume manufacturing facilities. We’ve brought a lot of innovation to the market, so we have compelling products (1055 patents filed in 2001). We have the broadest product portfolio. So, we’ve been committed, we stay committed, and our goal is to be the preeminent supplier of flash memory. We have a track record that proves we are going to be a force in the flash-memory market.
http://www.e-insite.net/commvergemag/index.asp?layout=articlePrint&articleID=CA214594
culater
Maury Wright, Editor-in-Chief -- 5/1/2002
CommVerge
At the edge of a converged network, you'll find a variety of intelligent devices that span applications from entertainment to productivity and reside everywhere from the auto to the living room. All of these nodes share some characteristics, such as connectivity and intelligence, and they all rely on some of the same key enabling technologies.
Nonvolatile memory is perhaps the most important of the enablers, even if processors get the most glory. Flash memory serves in cell phones, set-top boxes, digital music players, and a host of other devices, acting both as program storage and as a content store for music, pictures, contact lists, and many other data types.
Given the significance of flash, we decided to host a roundtable discussion on the topic. We felt that such a format might prove valuable because it would allow industry experts to pontificate on the issues directly. The summit took place only virtually—via email—but yielded a robust, realistic dialogue nonetheless.
Follow along to learn where flash-memory capacity and prices are headed, which applications will drive consumption, whether alternative nonvolatile memories will encroach upon flash markets, and other valuable insights.
CV: Because CommVerge generally focuses on convergence applications and uses that application-level focus to spotlight enabling technologies like memory, I'd like to start at the application level. Could each of you describe the three or four products that consume the largest quantities of flash memory today?
PARTICIPANTS
Philippe Berge, Director of Marketing, STMicroelectronics Memory Product Group
Bertrand Cambou, Group Vice President, Memory Group, Advanced Micro Devices
Keith Horn, Vice president of Marketing, Fujitsu Microelectronics
Bill Krenik, Wireless Advanced Architecture Manager, Texas Instruments
Brian Kumagai, Business Development Manager, Flash Products, Toshiba
Kevin Plouse, Vice President of Technical Marketing and Business Development, Memory Group, Advanced Micro Devices
Sudeep Sharma, Associate Vice President, Memory Division, Mitsubishi Electric and Electronics USA
Victor Tsai, Product Marketing Manager, Flash Products, Hitachi
Mike Williams, Director of Marketing, Flash Products Group, Intel
Bing Yeh, President and CEO, Silicon Storage Technology
Sudeep Sharma (Mitsubishi): Today the largest quantities of flash memory are consumed in cellular handsets, storage cards, BIOS flash applications for PCs, and portable electronic devices such as digital cameras, PDAs, and MP3 players.
Kevin Plouse (AMD): Cellular telephones use the bulk of flash memory today, and we don't expect that to change anytime soon. So, when we look in our crystal ball, we see a cell phone with more features that uses more flash memory. One key point is that the people who invested in 3G networks invested that money because those networks drive data, and they drive data to phones. The convergence of the phone and the handheld computer is the single largest opportunity for flash memory. The second driver is the consumer product. There are all kinds, but the ones that come to mind are music, video, and photo storage (cameras, video recorders, etc). If the cost is right, these will drive a lot of demand. Consumer appliances too—like DVD players, high-definition television—drive growing demand for flash devices.
Bertrand Cambou (AMD): And the third one is internetworking. Obviously, the dot-com explosion resulted in extensive investments in networks, and flash provided network reprogrammability. The dot-com explosion has been replaced with a dot-com collapse, but eventually the networks will be replaced. We don't know yet what they will look like, but we can assume that with the fundamental growth in data, we must keep data moving.
Mike Williams (Intel): Cellular phones by far consume the most flash in millions of megabytes. Cell phones are the highest volume, shipping approximately 400 million units with a large density mix. Digital cameras would be next, not because of their volume, but due to their higher average density. The next few applications—which include networking/communications, set-top boxes and handhelds—are all smaller and comparable in consumption.
CV: Today, the flash market is clearly divided into data- and code-storage segments, dominated by NAND/AND [not and/and] and NOR [not or] flash architectures, respectively. How do these different architectures match up with the flash applications? Also, please explain the need for flash in these applications and point out the key memory-system requirements, such as capacity, speed, cost, and others.
Brian Kumagai (Toshiba): Today's primary NAND applications are digital cameras (mainly in removable-card format), game consoles/accessories, and digital audio. Today's primary NOR applications are cell phones, PDAs, and set-top boxes. NOR applications are increasingly being restricted to code storage/execution, where the density requirements are relatively small, code must be executed from flash, and write performance/reliability are not concerns.
“The people who invested in 3G networks invested that money because those networks drive data, and they drive data to phones. The convergence of the phone and the handheld computer is the single largest opportunity for flash memory.”
Kevin Plouse,
Advanced Micro Devices
Mike Williams (Intel): Flash is critical to all these applications for supplying system and application code and data storage. The best way to split the flash market requirements is between code and code+data architectures/requirements (cellular, handheld, set-top, networking, and telematics) and pure data-only storage (digital cameras and MP3 players). Code and data applications typically require high-performance reads (burst or page-mode), read-while-write capability at 66-MHz, data integrity/reliability, low power, and mid- to high-density capacity, while data-only applications require and value high density in a removable form factor.
Philippe Berge (STMicroelectronics): In addition to mobile terminals (cell phones), we see PC BIOS, automotive, and digital home gateways as key markets. In mobile terminals, the key requirements are low power consumption, high-density, tiny packages and footprint, an optimized interface with the baseband processor, and the ability to combine flash with SRAM. These nonvolatile memory requirements are directly driven by more and more user-friendly application features, such as GPRS [general packet radio services] and WAP [wireless application protocol], Internet and talk-mode protocols, tri-band support, voice memos, voice recognition, predictive text input, and color displays.
Games require bigger and bigger operating systems, hence bigger and bigger nonvolatile memory that has to be executed as fast as possible. Longer standby and talk times require low-power supply and low-power operative and standby consumption. In digital home gateways, the key requirements are cost and write and programming throughput. They are driven by the following system evolution: Web navigation, e-commerce, expert systems for user profiling, and remote software downloads for things like operating-system updates and TV program guides.
CV: From a pure silicon perspective, discuss your organization's technology roadmap in the NAND and/or NOR camps. Tell me where you stand today in terms of capacity and where you expect to be in 2005. Please describe, at a high level, the techniques that will deliver on your roadmap, such as multilevel cells (MLCs) as opposed to single-level cells (SLCs).
"Until a product is proven in production with real customers, it is difficult to place much faith in it. MRAM has been researched for over 30 years, yet it is still not in mass production."
Bing Yeh, Silicon Storage Technology
Bing Yeh (Silicon Storage Technology): SST is aligned in the NOR camp, and we believe this will continue to be the dominant area for flash, especially for code storage, but also for data storage. Code storage requires fast access times for system boot-up and reliable byte access without the latency that is common in NAND flash. Furthermore, in low density, NAND cannot compete, as it requires massive overhead circuitry to implement.
MLC technology will bridge the gap in cost between NAND and NOR in the medium and higher densities, and we foresee a realistic roadmap to four bits per cell using MLC SuperFlash technology. Currently, SST has a wide range of capacity in the low densities from 256 kbits to 16 Mbits. We will expand into the medium densities, from 32 Mbits and up, for the coming years for the code-storage applications. We also plan to offer more than 1 Gbit per chip for the mass data-storage applications.
Sudeep Sharma (Mitsubishi): We are primarily focused on DINOR technology, a special type of NOR architecture. Relative to NOR flash, our DINOR technology offers faster random access at a lower voltage, and seven to 25 times quicker erase cycle. All of our flash-memory parts also have a BGO (background operation) function. Mitsubishi was the first to adopt the BGO function in 1997 on 8-Mbit flash. BGO can eliminate EEPROM [electrically erasable programmable read-only memory] from cellular phones since data can be read from banks while another bank is being programmed or erased.
Mike Williams (Intel): Our product portfolio is focused on NOR, not only for code but also for specifically optimized code+data requirements. We have three product lines. Our high-performance Wireless Flash, for handheld customers requiring the ultimate performance, offers a 1.8-volt (3-volt I/O option) product family with densities from 32 to 128 Mbits. Currently in production on 0.18-micron technology, we are sampling now on 0.13-micron, with a roadmap to 90 nanometers. Also on 0.13-micron, we are adding a new x32 implementation and increasing density to 512 Mbits by 2005.
Intel StrataFlash Memory is the highest-density, lowest-cost flash memory for code+data applications. Used in nearly every WinCE/PocketPC handheld, today's StrataFlash memory on 0.18-micron is Intel's third generation of MLC technology, which we originally introduced in 1997. StrataFlash is offered today in 32- to 128-Mbit densities and a 256-Mbit density later this year at 3 volts (1.8-volt I/O available). A high-performance 1.8-volt version will be released later this year, and densities on the MLC technology will reach 512 Mbits by 2005.
“The anticipated largest consumers by 2005 should be cell phones, consumer electronics (digital cameras, MP3 players), networking, and automotive (including engine control and navigation systems).”
Keith Horn, Fujitsu
Industry-standard boot block (C3/B3) flash, now in its fourth generation of complete backward compatibility, is currently in production on a 0.13-micron process. This product family includes 8 to 64 Mbits, and production will continue through 2005 and beyond. In addition to continuous improvement, leading lithographies that keep us one product generation ahead of our nearest competitors, and proven multilevel cell manufacturing, we are exploring the use of four bits per cell and Ovonyx Unified Memory to expand our roadmap in the coming years.
Brian Kumagai (Toshiba): In NAND flash, we are currently in mass production of 512-Mbit monolithic SLC, 1-Gbit stacked (two-chip) SLC, and 1-Gbit monolithic MLC. In 2005, maximum density will increase to 4-Gbit SLC and 8-Gbit MLC monolithic devices. In NOR flash, our highest density today is a 128-Mbit SLC. We have plans for 256-Mbit and possibly 512-Mbit MLC devices in 2005.
Kevin Plouse (AMD): FASL [Fujitsu AMD Semiconductor Limited, AMD's joint venture with Fujitsu] is a leader in NOR technology. We are neck-and-neck between FASL and Intel for first and second position in the market. The NAND/NOR line is getting more and more blurry in terms of applications. Our customers prefer NOR but they want cost reductions.
Bertrand Cambou (AMD): One path to cost reduction is MLC.... [However,] we don't see how 4-bit MLCs can work reliably for code-storage solutions. As a result, we see the classical floating-gate technology coming to a point where it is not extendable anymore. That is why AMD took a different path with our MirrorBit architecture, which is not based on the MLC principle.... For years we have worked to develop an alternative path and now we are working full speed on MirrorBit—a technology without the compromises associated with MLC. We also recognize that MirrorBit is very expandable, even to four bits per cell. We believe that the move away from floating-gate will happen and our conviction is strong that we are engaged in a paradigm shift.
“There will be a sustainable need for code-storage flash that will be driven by the need for bigger and bigger operating systems enabling more and more user-friendly applications.”
Philippe Berge,
STMicroelectronics
Victor Tsai (Hitachi): We are a major supplier of data-storage flash with our MLC AND-type flash technology, and we are a manufacturer of code-storage NOR flash products. Hitachi recently introduced the new AG-AND multilevel flash memory cell, which gives Hitachi a technology and cost advantage over competing data-storage flash products and technologies.
Keith Horn (Fujitsu): We currently offer only NOR flash. However, our Multi Chip Package lineup will continue to provide both NOR and NAND flash. The company's flash roadmap offers an impressive range of densities (2 to 128 Mbits), voltages (5 to 1.8 volts), and we have a well-established reputation for advanced packaging methods.
CV: Is there the possibility that the flash industry might consolidate toward a single type of flash architecture? For example, could NAND flash be augmented with DRAM cache and control circuits that would allow code-storage applications to leverage the low-cost, high-density benefits of NAND flash? Or, are there breakthroughs in the NOR world that can ramp capacities and lower costs to compete with NAND flash?
Keith Horn (Fujitsu): The disadvantage of NAND flash is its reliability. Some applications simply cannot risk reliability issues and will be forced to continue to utilize NOR or NOR-like flash. However, the production of multibit cell flash product will allow higher NOR-like reliability with pricing that is more in line with NAND flash.
Bing Yeh (Silicon Storage Technology): NAND- and NOR-type applications and specs are quite different, and both types will coexist forever. Four-bit-per-cell MLC will provide a great challenge to NAND flash in terms of cost. However, because several large Japanese companies have focused on NAND flash, there will continue to be some NAND flash market inertia. So, regardless of what arguments technologists might make about whether NOR or NAND is technologically better, NAND will continue to play a role in the high-density flash market. In the embedded and mainstream code-storage markets, however, NAND will never penetrate. We see clear evidence of this, as NAND vendors have followed a DRAM model in the manufacture of NAND flash, pushing products into higher and higher density and not even offering NAND devices anywhere below 64 Mbits, which is the domain of code storage today.
Victor Tsai (Hitachi): There may be a point in time where there would be a convergence of data-storage and code-storage flash. Data-storage flash is generally more cost-effective than code-storage flash. Hitachi has just introduced the superAND flash product, which incorporates some NOR-like features, including power-on read for system boot-up and 100 percent good memory without error handling and memory management by the host CPU. This is the first crossover product that can satisfy both data-storage and code-storage needs in a system.
Sudeep Sharma (Mitsubishi): We don't see the NOR and NAND types of flash-memory architectures converging. However, new flash-memory architectures may be developed to handle both types of applications.
Mike Williams (Intel): We believe it will continue to fragment. Application requirements are diverging rather than converging. We see this today in multiple line-item offerings on our silicon and numerous stacking configurations requested. Additionally, our long-term strategic alignment with our top customers indicates continued diversification. One size certainly does not fit all.
"Application requirements are diverging rather than converging. Our long-term strategic alignment with our top customers indicates continued diversification. One size certainly does not fit all.”
Mike Williams, Intel
And let me correct a potential misconception with our Intel StrataFlash memory on leading-edge lithographies. We believe we do compete with NAND on a cost basis. The question isn't about cost per se, but about what price a manufacturer is willing to sell that flash device. Currently, NAND manufacturers are pursuing a very aggressive pricing strategy to make up for what we believe is an inherent mismatch with the system requirements in a code or code+data environment (bad blocks, error correction, read speeds, increased system memory, etc).
CV: Mike, is it your point that NAND manufacturers have cut prices to artificially low levels to gain entry into code or code+data storage applications, and that some buyers will deal with mismatched characteristics like slow read speeds to buy the lower-cost flash? And when you say you "compete with NAND on a cost basis," are you making that claim based on system costs in a code or code+data application?
Mike Williams (Intel): There were some lofty expectations for NAND growth the past few years, mainly driven by growth projections for digital cameras and digital music players. Each year, the forecasts were pushed out another year. The missed growth expectations have left the NAND suppliers scrambling to find homes for their products, and they have resorted to trying to fit their products into the traditional NOR markets. But NAND has inherent feature mismatches for these applications. For example, you cannot execute out of NAND, given the slow read speeds. Therefore, redundant memory, consuming more space and power, is required for the device to operate. NAND also requires error-correction circuitry. NAND contains bad blocks that must be managed. And the list goes on.
"In the near-term, the 'perfect' memory, nonvolatile RAM, will remain an R&D product. While some technologies appear promising, we believe the applications will be restricted."
Brian Kumagai, Toshiba
In short, there are a number of system-complexity issues when designing with NAND, and the NAND suppliers are attempting to overcome these issues by using cost as an incentive. Hence, NAND is selling at a very aggressive price today. In most cases involving both code and code+data applications, Intel StrataFlash memory offers a lower overall system cost and is much easier to use in design.
CV: Given the state of the market today, and developments in NAND and NOR technologies, take a look at your crystal ball and project the top three or four products for 2005. And again, please explain the key memory-system requirements for flash.
Keith Horn (Fujitsu): The anticipated largest consumers by 2005 should be cell phones, consumer electronics (digital cameras, MP3 players), networking, automotive (including engine control and navigation systems). As cell phones offer more and more features, they will continue to require larger densities of memories and smaller packages.
Mike Williams (Intel): By 2005, cellular, cameras, networking, PDAs, and set-top boxes will remain as the top markets, in our opinion, with cellular continuing to lead and handheld growth most likely outpacing the others. Telematics/GPS will also emerge as a top flash application.
Brian Kumagai (Toshiba): In 2005, NAND applications will include digital still/video cameras, cell phones (mainly for digital camera/audio/video purposes), PDAs, and set-top boxes. In all of these applications, whether the flash is used for code and/or data storage, the primary factors driving the usage of NAND are the requirements for high density and low cost. Additionally, for the data-intensive applications, the superior write performance and reliability of NAND compared with NOR is an important consideration. NOR applications for 2005 include cell phones, low-end set-top boxes, and networking/communications equipment.
Kevin Plouse (AMD): Looking into our crystal ball though, we can't forget to talk about the auto dashboard. It's small, but the fastest growing forecast is in the cockpit of the car—for entertainment and navigation. The car PC has just hit the inflection point for growth. It's been in development for eight years or more and is now becoming a standard part of the car.
“The car is a very interesting environment for us because we have been focusing on the car for a while, and you do not use substandard flash in a car.”
Bertrand Cambou,
Advanced Micro Devices
Bertrand Cambou (AMD): The car is a very interesting environment for us because we have been focusing on the car for a while, and you do not use substandard flash in a car (for example, because of the extreme temperature variation requirements). And that has been our strength.
CV: With cell phones identified as such a huge consumer of flash memory, could you further illuminate how flash is used in those devices. Digital cell phones rely on high-speed DSPs, so I know SRAM is required for code execution. Perhaps you could provide a scenario for what types of code and data are stored in different memory types, both when a phone is standing by and when a call is in progress. And describe how close this model of memory usage comes to other applications like PDAs or telematics systems.
Mike Williams (Intel): Flash memory has traditionally been used in cellular handsets to store program code used to control the operation of the device, to store data for device-tuning parameters, and to store data such as frequently used phone numbers and other personal information. Flash was adopted in these devices due to its solid-state ruggedness and high data retention—a phone can be dropped to the floor, the battery can be removed, and the information in the flash memory is retained.
Internet capable handsets, including new 2.5G and 3G phones, are driving the requirements for higher-performance and higher-density flash memory. These cellular handsets can be separated into two main processing functions: the baseband communications processor and the applications processor. Flash memory is used in the baseband unit to store program code for the traditional microcontroller device in charge of handling the specific cellular protocol. Flash memory can also be used in the baseband unit to store the DSP algorithms, as well as acting as the main memory in the event of an onboard cache miss from the integrated SRAM memory. Regardless of standby operation or active operation, the baseband processor is continually executing code from the flash device. In standby mode, approximately 1 to 3 percent of the time (depending on the actual protocol), the baseband processor must "wake up" to ping the nearest basestation in order to stay connected.
Flash-memory requirements are exploding on the application-processor side, where flash is used to store program code for new functionality such as Web browsers, color displays, Java applets, and audio/digital data manipulation. Connecting to the Internet opens up the need for more data on the application-processor side for storing large video files, digital music files, photographs, and email.
The memory usage between an application processor in a cell phone and a PDA are the same (hence the convergence of cell phones and PDAs). The industry debates whether one common multipurpose device will emerge or whether we will continue to see a variety of devices tailored for a specific need. Whether a cell phone, PDA, or telematics, Intel is offering common building blocks, including baseband processors, applications processors, and flash memory based on the Intel Personal Internet Client Architecture—a development blueprint for wireless devices and software combining voice and data.
CV: Brian Kumagai of Toshiba seems to imply that future cell phones will have a mix of NAND and NOR flash. I assume the former will serve integrated add-on functionality like a digital camera or MP3 player, while the latter, I assume, serves to store code for the cell-phone application. Could you give me a precise picture of how you see this memory architecture applied?
“Although in many ways integration is the key to low cost, we believe that discrete flash memory will continue to be less expensive than embedded flash memory on a per-bit basis.”
Sudeep Sharma, Mitsubishi
Brian Kumagai (Toshiba): We expect both evolutionary and revolutionary cell-phone architectures utilizing NAND flash. The evolutionary architecture will utilize NAND for data (photos, audio, video, etc) and the NOR for all types of code storage. In this case, the NOR will have to be fast enough to support code execution for all processing/control functions, including the DSP, which will probably be realized by page/burst mode. Toshiba would expect the NOR density to increase at about its historical rate for this architecture, since the NAND will take over some of the previous NOR functions, such as phone-number storage. The revolutionary architecture would use only NAND combined with lots of RAM (probably DRAM). In this case, the NAND would store all of the code and data, and the code would be executed out of RAM. The smartphone and PDA-combo phone will drive the transition to the NAND-only architecture.
CV: Outside of external CompactFlash, SmartMedia, SD Card, and Memory Stick modules, will there be a sustainable need for stand-alone flash memory chips going forward? Integration is the key to low cost, and SOC [system-on-chip] is an unmistakable trend. Will flash become largely a feature integrated onto other chips? If not, describe the capacity requirements or silicon limitations that will prevent such consolidation.
Philippe Berge (STMicroelectronics): There will be a sustainable need for code-storage flash that will be driven by the need for bigger and bigger operating systems enabling more and more user-friendly applications. Embedding flash will always remain a tradeoff of cost, footprint, and performance. Overall, flash will keep growing in three directions: standalone, embedded, and cards. Standalone and embedded flash will grow mostly for code storage. Flash cards or flash-plus-other-memory cards will develop as real subsystems for data storage. There is no real standard package yet, but the evolution of a standard associated to cost/bit reduction will push the market to a higher level of volumes/value.
Sudeep Sharma (Mitsubishi): Flash memory is being integrated already and will continue to be more integrated in SOC devices. However, the density of flash memory in SOC applications will continue to be limited because of chip-size constraints, which is also related to the yield issue. Continued development of finer process technologies will increase SOC flash density, but future applications will also continue to demand more flash memory density. Although in many ways integration is the key to low cost, we believe that discrete flash memory will continue to be less expensive than embedded flash memory on a per-bit basis.
Bill Krenik (Texas Instruments): Texas Instruments doesn't make flash, but as a leading vendor of ICs for cellular handsets we have a vested interest in flash developments. For wireless, integrating flash memory on the same chip may result in an inflexible memory configuration, because the handset designer will need to specify the amount of flash memory to be integrated early in the design cycle, before actual memory needs are clear. This may result in excess memory, leading to a cost penalty, or insufficient memory, resulting in loss of product features or the need for added external memory.
Further, flash integration normally requires six to eight additional process reticles over a conventional digital CMOS [complementary metal oxide semiconductor] process, significantly increasing manufacturing costs. Since there are no significant performance benefits obtained by integrating flash onto the same chip for wireless, it is difficult to justify flash integration. Other options, however, such as the use of multidie packaging, may be attractive in some cases.
Bing Yeh (Silicon Storage Technology): SST is by far the world's leading, if not the only substantial vendor, of embedded-flash solutions. Our split-gate SuperFlash architecture is available as integratable intellectual property through many of the world's leading foundries. Today, dozens of blue-chip companies license SST's SuperFlash technology for integration into their own wireless chips and other ASICs on a regular basis. Since SuperFlash technology is CMOS-compatible for fab portability and scalability, and since SuperFlash offers significantly better power usage and die size efficiencies compared with both stacked-gate and NAND flash, SST believes this will continue to be a rapidly growing and successful market.
That being said, however, flash is not always as cost-effectively integrated with other system requirements onto a single chip, due to the additional processing and testing steps required for flash memory. SOC with flash is most effective on very small die, where most of the component cost is in the packaging, or at the very high-density spectrum of so-called smart flash memory, where most of the SOC silicon is occupied by flash.
CV: The mixture of memory technologies on one dedicated memory chip or in a multichip package appears to be another trend in the integration story, and without doubt many convergence products require a mix of memory types. Give me your opinions on what types of mixes will be popular, including the possibility of mixing multiple flash types along with SRAM and DRAM. Describe why a chip dedicated to a mix of memory is a good idea, and if so, how you can craft a standard product family that meets the needs of different applications.
Mike Williams (Intel): Providing one packaged memory subsystem is compelling for handheld devices due to the space savings. Today, we are stacking flash and SRAM into one package, and the possibilities are almost endless for stacking, including flash and flash, flash and logic, flash and other memory, and any of the combinations above. These combinations are driven by the memory-subsystem needs of our customers. Crafting one standard product family is not achievable due to the fragmentation discussed previously. Successful flash suppliers must strive for flexibility and quick turnaround time to meet their customers' specific needs.
Keith Horn (Fujitsu): Our lineup of multichip packaged (MCP) devices, which includes flash and SRAM or flash and Fast-Cycle RAM, will continue to lead the field in mixed-memory technology on one package. For cellular applications, this MCP device can replace multiple components, resulting in space savings. It can also offer higher densities that are not available in today's marketplace with a one-chip solution and at a reasonable cost.
"Rotating storage is not a practical memory solution for today's handsets. However, in the future, the technology may be a good fit for high-end PDAs."
Bill Krenik, Texas Instruments
Bill Krenik (Texas Instruments): In wireless-handset applications, SRAM is normally used for multiple levels of cache, while flash is used for program storage and storage of user data and system settings. Since the cache needs to be integrated with the processor and flash integration appears to be cost prohibitive for wireless, it seems unlikely that SRAM+flash products will emerge.
CV: Do you see any near-term prospects for technologies like MRAM [magnetic RAM], FRAM [ferroelectric RAM], Ovonyx's optical technology, or some other nonvolatile memory to succeed in mainstream applications? Also, can rotating storage technologies, like hard-disk drives and the new Dataplay drive, impact the market for flash modules?
Bing Yeh (Silicon Storage Technology): Until a product is proven in production with real customers, it is difficult to place much faith in it. MRAM has been researched for over 30 years, yet it is still not in mass production.
Brian Kumagai (Toshiba): In the near-term, the "perfect" memory, nonvolatile RAM, will remain an R&D product. While some technologies appear promising, we believe the applications will be restricted. For example, Toshiba is developing 32-Mbit and 64-Mbit FeRAM [FRAM], which can be used to replace NOR+SRAM in low-end cell phones. Still, none of these new technologies will reach the density or cost-per-bit of NAND flash. Toshiba plans to introduce a commercial FRAM by the end of 2002. The target density is 32 Mbits. The primary technical challenge is acceptable performance in terms of access time.
Bill Krenik (Texas Instruments): Of the advanced memory technologies you cite, only FRAM is proven in high-volume manufacturing today. FRAM is also attractive because it can be integrated with very few additional process reticles. While MRAM and Ovonyx memory are very interesting technologies, they remain unproven as real solutions for low-cost, high-volume applications.
Rotating storage, of course, looks great on a cost-per-bit basis. However, this low cost is only available for relatively large memories. As a result, rotating storage is not a practical memory solution for today's handsets. However, in the future, the technology may be a good fit for high-end PDAs.
Mike Williams (Intel): As we've discussed publicly, Intel is pursuing Ovonyx memory technology. Although it is still early in the development, the initial results look encouraging. Compared to MRAM and FRAM, we believe Ovonyx holds the best promise for delivering on the performance, densities, integration, and reliability needed for our customers. If all goes well, we would expect Ovonyx to start making an impact on mainstream applications as early as the middle of the decade. But it is premature to discuss specific product plans.
Rotating storage technology will always be an alternative in the pure data-storage area. We see this in the digital-music-player market segment today, where NAND memory is being squeezed by less-expensive rotating technologies.
CV: We’ll finish with the unpopular question. I’d welcome your views on where flash prices are headed. I’d like to discuss price for two reasons. First, low price enables convergence applications. Music players, for example, have been hampered by flash prices, although I know they’ve dropped considerably (and I know the RIAA [Recording Industry Association of America] has hit the music players harder, but that’s a discussion for another day). However, low price has potentially negative ramifications for flash manufacturers. Moreover, the number of manufacturers making flash today is still relatively large relative to other commodity memory types like DRAM. Is the flash market headed for a major consolidation toward a small group of major players? What characteristics of your business make you a long-term participant in the flash industry?
Bertrand Cambou (AMD): It is our belief and strategy that what we need is to continuously and relentlessly cut the price-per-bit. And to commit to our customers a cost reduction that empowers them to build higher and higher densities into their systems, thereby making flash even more pervasive.
Philippe Berge (STMicroelectronics): We are expecting prices to stabilize in Q2 due to the recovery of the demand and to rise in the second half of the year. As for consolidation, the high end of the code-storage market is already at an advanced stage, with very few suppliers having the proper relationship with the key customers, advanced technology, and manufacturing capacity. High-density flash devices are already coming from only three to four suppliers. Second-tier vendors are shipping devices made with lower-density older technology. In the long term, flash technology is essential for STMicroelectronics’ SOC strategy. Flash offers ST the advantage of both differentiated and standard products. The flash-differentiated products, essentially custom configurations for high-volume applications, are key for our corporate strategic customers and give some stability to the business. The standard-product portfolio contributes by extending our customer base and providing the volume to lower our overall manufacturing costs.
Victor Tsai (Hitachi): There are many code-storage NOR flash suppliers, but the number of data-storage NAND/AND flash suppliers is relatively small. The growth rate of data-storage flash is much higher than code-storage flash, so while there may be consolidation in the code-storage flash supplier base, there is still a lot of market potential for new entries into the data-storage flash market.
Sudeep Sharma (Mitsubishi): We believe flash-memory demand will increase strongly and likely outstrip supply as the US and worldwide economy improves and as cellular handset demand increases. Mitsubishi Electric has been strong for a long time in providing a wide variety of memory technologies that can be combined to provide a complete solution.
Keith Horn (Fujitsu): Flash prices appear now to have stabilized. We have not seen prices increase yet, but they certainly are not decreasing. Low prices may eliminate some newcomers to the flash business, but established flash manufacturers will continue to thrive by implementing die shrinks and investing in new technologies such as multibit cell product. Fujitsu should be considered a long-term participant because of our joint venture with AMD and the considerable investment that has been made in our facilities.
Kevin Plouse (AMD): Flash has attracted every major memory player. Those that are the strongest will survive, the best technology will survive, the most innovative will survive. We’ve been a leader in nonvolatile memory for more than a quarter century. We’ve built a strong partnership with Fujitsu. With Fujitsu, we believe we have the best high-volume manufacturing facilities. We’ve brought a lot of innovation to the market, so we have compelling products (1055 patents filed in 2001). We have the broadest product portfolio. So, we’ve been committed, we stay committed, and our goal is to be the preeminent supplier of flash memory. We have a track record that proves we are going to be a force in the flash-memory market.
http://www.e-insite.net/commvergemag/index.asp?layout=articlePrint&articleID=CA214594
culater
Join the InvestorsHub Community
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
