Depending on the system application, the dominant memory type will change as system requirements change. The memory market is generally split into three main product types: DRAMs, SRAMs, and nonvolatile devices. For the most part, memory manufacturers strive to achieve three main goals. These are shorter access times, higher capacities, and lower power consumption. As a result, they're developing higher-performance bus interfaces, leveraging smaller feature sizes and process improvements to build faster transistors, and scaling the supply voltage to reduce power.
In computing, for example, the need for large amounts of temporary storage places the emphasis on DRAMs. Today, 256-Mbit commodity 133-MHz synchronous DRAMs (SDRAMs) are giving way to first-generation double-data-rate (DDR) SDRAMs. These currently come in speed grades of 266 and 333 MHz. First samples of 400-MHz, 256-Mbit DDR SDRAMs are just starting to hit distributors, and 512-Mbit devices won't be far behind. Expect the main supply of 400-MHz and faster devices (possibly up to 667 MHz) to arrive in late 2003, when the second-generation DDR interface (DDR II) goes into production. Specifications for the third-generation DDR interface are almost done, providing a transition from the DDR II interface starting at 667 MHz. DDR III DRAMs probably won't begin sampling until mid- to late 2004.
For data communications and networking, static RAMs get the most play. They traditionally offer much faster access times with little or no latency. Specialized SRAMs with no bus-turnaround latency (some companies call this NoBL and others call it ZBT for zero bus turnaround) and other devices with separate input and output ports (quad data rate) are pushing to deliver data faster. New memory approaches, like the SigmaRAM, are also involved in the faster data delivery game. But DRAMs aren't falling behind with the reduced-latency DRAM, the fast-cycle DRAM, and next-generation Rambus RDRAMs emerging to meet the performance demands of these systems.
Consumer applications ranging from cell phones to digital cameras and MP3 music appliances turn the tables on DRAMs and SRAMs. Although many of these systems may contain some limited amount of DRAM or SRAM, they typically consume lots of flash memory. The densest NAND-based flash devices now being sampled are 2-Gbit chips. By late 2003, 4-Gbit devices should be coming out of the research labs. New nonvolatile devices based on ferroelectric materials are starting to appear at density levels of 16 and 64 Mbits per chip. On the drawing boards is yet another magnetically based nonvolatile memory known as the magnetoresistive RAM (MRAM). Though still experimental, these devices hold a lot of promise for the nonvolatile storage market.
TOP TEN >512-MBIT SDRAMS will go into volume production with data-transfer speeds of 333 MHz.>FERROELECTRIC RAMS featuring 64-Mbit capacities will see limited production in 2003.
>VOLUME PRODUCTION OF 2-GBIT NAND-BASED flash memories is expected for this year.
>NOVEL DATA DOUBLING TECHNOLOGIES, such as multilevel charge storage developed by Intel, and the mirror-bit scheme introduced by Advanced Micro Devices, promise samples of 4-Gbit flash devices by late 2003 or early 2004.
>FIRST PRODUCTION SAMPLES of the high-speed 36-Mbit SigmaRAM will be released by mid-2003. These will Include 333- and 400-Mbit/s data-transfer speeds for single-data-rate interfaces, and speeds of 666 to 800 Mbits/s for double-data-rate (DDR) implementations.
>SAMPLES OF THE HIGH-SPEED DDR II SDRAMS with 400-MHz data transfer rates will bubble to the surface from several suppliers. By 2004 these devices will be the mainstream memory for PCs.
>REDUCED-LATENCY DRAMS at 256-Mbit densities are due for release this year. Data transfer rates of 600 Mbits/pin will be possible, along with a cycle time of just 25 ns.
>SPECIFICATIONS FOR THE SECOND-GENERATION RLDRAM will be finalized, with samples of the devices slated for 2004. Data rates will hit 800 Mbits/pin for the RLDRAM II, and cycle times will drop to 20 ns. This will let the device compete head-to-head with both RDRAMs and most of the high-performance static RAMs.
>SECOND-GENERATION FAST-CYCLE DRAMS (FCRAMs) should go into limited production with capacities of 288 Mbits, and samples of 512-Mbit first-generation FCRAMs are targeted for late 2003. In addition, specifications for the third-generation FCRAMs will be finalized, with samples projected for late 2004.
>LIMITED PRODUCTION OF NEXT-GENERATION 256-MBIT RAMBUS RDRAMS is on tap for this year. Their simplified internal bank architecture will let systems operate at lower power levels and compete more directly with SDRAMs on a price-per-bit basis.