The Modern Era

---

Today, there are two major kinds of random access (i.e., any part can in principle be accessed without major differences in time to access) computer memory:

• dynamic RAM (DRAM) -- DRAM is designed for relatively low cost rather than highest possible speed. DRAM uses a transistor and a capacitor to store a bit (a 1 or a zero -- a binary digit). The capacitor has to have its charge refreshed. DRAM is used for the main memory of current computers
• static RAM (SRAM) -- static RAM is designed more for high speed than for low cost. Each bit is stored using 4 to 6 transistors, so SRAM is bigger (or looked at the other way, the same sized chip stores less) than DRAM. SRAM is consequently more expensive. SRAM is used in current computers to implement caches. A cache is a small, high-speed memory, that is designed to hold information likely to be frequently accessed, thereby hiding the effect of a slower memory

There are many attempts at bridging the gap between the low cost of DRAM and the high speed of SRAM. Two examples:

• RamBus -- some cache is included on the DRAM chip, and a high-speed interface to the whole memory system is included as part of the design
• EDO RAM -- if a future access happens immediately after the position where a previous access occurred, the future access needs less setup time and is a lot faster

The latest generation of DRAM stores 64Mbits per chip; around 2002, we should expect to see chips with 1Gbit per chip. Here is a detail of some stages in the manufacture of a 1Gbit chip, containing parts as small as 0.2 microns. The charge on individual capacitors is about 28.5fF (the "f" stands for femto, 10-15). The next DRAM generation, the 512Mbit chip, should appear around 1999. By the time the Gbit DRAM generation arrives, memory access time will be about 30ns, about twice as fast as the fastest DRAM available today. Chip sizes are generally given in terms of bits. Historically, a single bit was obtained from one chip, and several chips were used at once to access bigger units such as bytes simultaneously. Today, chips are available which allow access to more than one bit at a time, but the terminology persists. The work illustrated here was done by NEC in Japan in 1994, illustrating the long lead times needed to get advanced new technologies out.

Memories are Made of This 17:30 Thurs 12 September 1996 SHB5, Senate House, University of the Witwatersrand, Johannesburg, South Africa