Digital ICs> Memory Markets Gather Momentum In 2004

While the PC market remained the catalyst for the memory market's growth in 2003, other applications will begin to create some noise. DRAM will further expand into PDAs, smart phones, digital video recorders, advanced gaming consoles, and VoIP infrastructure. The amount of functionality and data that future PDAs and smart phones will provide requires more buffering, necessitating increased memory usage. And as digital-video-recorder (DVR) capabilities are incorporated into more set-top boxes, these systems will require larger buffers and more temporary storage for the digital content, resulting in a need for higher density memory.

Advanced video game consoles, namely the forthcoming Xbox 2 and Playstation 3, will push beyond the current 64 Mbytes of memory employed in current-generation systems to at least 256 Mbytes. This isn't a stretch, considering many of the high-end graphics cards that already come with 256 Mbytes. Also, the increasing use of voice over Internet protocol (VoIP) as the standard communications platform will require more memory in everything from the handset to the central office.

While the introduction of 64-bit computing began in 2003, it won't have an immediate effect on the computing market. However, the ability to address more than 4 Gbytes will have more implications for future computing. Many applications could benefit from such large memory spaces.

The SRAM market is poised for a positive 2004 too, with the fourth quarter of 2003 bringing rising prices and the promise of device shortages. Price increases will then occur during the first half of 2004. Semico Research Corp. expects this year to show a growth of 15.6% in revenues to $2.9 billion and a 6.8% growth in units to 1.1 billion.

For 2005, Semico forecasts an overall market downturn. The SRAM market will follow suit with a 20.2% decline in revenues to $2.3 billion and a 13.3% decline in units to 963.2 million. Semico doesn't expect a long-term recovery of the SRAM market, due to the influence of DRAM and embedded memories on SRAM's bread-and-butter markets. By 2008, we see unit shipments down to 896.4 million and revenues at $2.6 billion.

Demand is particularly high for asynchronous SRAMs and pseudo-SRAMs (PSRAMs) for use in cell phones. Such low-power devices are needed in higher densities for the increasingly popular camera phones with color screens. PSRAMs have made quite an impact in the market, dramatically lowering the aggregate ASP this year. They have a one-transistor cell structure like DRAMs and the low-power asynchronous interface of SRAMs. This makes them ideal from a power, cost, and density standpoint for cell phones. Semico expects PSRAM to make up 56% of the SRAM market in terms of units by 2008, growing from 11% in 2004.

We also see the beginning signs of a turnaround on the very fast, synchronous side of the SRAM market. After a couple years of stagnation, 2004 should bring a strong market for next-generation, higher-density synchronous SRAMs, such as the quad-data-rate SRAMs offered by Cypress, Renesas, IDT, NEC, and Samsung. The higher-density parts, from 9 to 36 Mbits, target the next generation of networking equipment.

In the flash market, a turnaround has already occurred. NAND flash has been in a shortage since May 2003, and the NOR market reached a market balance last August. In fact, at $11.3 billion, 2003's overall flash market (NAND + NOR) should be larger in 2004 than it has ever been.

Manufacturers are responding in a big way. Samsung converted massive capacity from DRAM to NAND and is now converting much of that to 90 nm. Toshiba and its partner SanDisk are boosting capacity by using multilevel-cell (MLC) technology and process shrinks. Renesas (formerly Hitachi) is making more room for its AND flash. Infineon and Hynix (along with partner STMicroelectronics) are sampling their own NAND flash products.

NOR manufacturers are a few months behind NAND vendors in announcing massive capacity additions. Expect to see many similar announcements this year. Intel has a number of fabs that could convert to flash in a relatively short time. Spansion (formerly AMD's and Fujitsu's FASL joint venture) hasn't announced anything yet, but AMD added the "Fab 25" in Austin, Texas, to its array of flash fabs. STMicroelectronics is likely to respond to the shortage with its numerous leading-edge fabs. Even SST recently announced a deal with PowerChip of Taiwan to develop flash chips based on PowerChip's 110-nm process and 300-mm wafers. The companies hope to scale those devices to 65 nm over the next few years.

Semico expects to see the current shortages continue through much of 2004. What's unknown is how rapidly DRAM makers will be able to respond to the new need for NAND. If they're successful, and if process migrations continue to move at their current unprecedented speed, then watch out for NAND to reenter an overcapacity situation by the middle of 2004. As for NOR, Semico expects a shortage to last into 2005.

However, considerable product development is taking place to find new devices that will replace SRAM. Based on DRAM, MRAM, FRAM, or FCRAM, these new devices, along with an increase in the use of embedded SRAM, explain the decline that we foresee in the SRAM market. Some of these devices, though, won't be on the market in full production for another six to 24 months.