Editorial

Embedded DRAMs Ride the Confluence of Technologies With the convergence of process technology and intellectual property, designers can build systems on a chip, but they still have plenty of gates left over to exploit. Enter embedded DRAMs. by Jonah McLeod

I'm a great believer in the enabling power flowing from the confluence of different technologies. Let's go back, for example, to 1979. Intel had just announced the 16-bit 8086 and its 8-bit cousin, the 8088. Shugart Technology had just introduced the 5.25-in. floppy disk drive. The combination of the two produced the PC revolution that continues to this day.

We're going through just such a technology confluence right now: the combination of process technology and intellectual property, which is sweeping electronics design into the system-on-a-chip era. Semiconductor process technology has made possible chips containing several million gates, which in turn give designers the capacity for creating complete systems on a single die. IP, which puts complex system functions in designers' hands, exploits the large gate counts while delivering the productivity boost that makes system-on-a-chip design practical.

Despite the use of IP, such as processor, DSP, and bus interface cores, designers are having a tough time completely filling these large-gate-count chips, even when they embed several cores. For example, in IBM and Motorola's latest implementation of the Power-PC 603e microprocessor, built with 0.25-µm process technology, the entire device is contained in a mere 47 square millimeters (taking die size as an approximate equivalent of gate count). Thus this powerful CPU would occupy a small percentage of a typical die of some 500 square millimeters.

However, one piece of intellectual property offers a solution for all that unused die: DRAM could easily fill the unused capacity. Take the new 3D and video graphics controller chips coming on the market. A 0.35-µm process will allow 3 Mbytes of on-board DRAM for the graphics display memory. The next-generation, 0.25-µm processes will allow a full 4 Mbytes of DRAM on-chip.

Furthermore, the rush to provide embedded DRAM is fostering a plethora of cross-licensing arrangements between DRAM suppliers and vertically integrated semiconductor vendors lacking DRAM technology. But these deals represent the initial way of building a system on a chip, in which one IC vendor supplies all the IP needed to build the circuit.

In today's free-for-all market, a new way has already emerged: start-ups appear to sell IP, and small design shops offer their design work as commercial cores. In this free-for-all, embedded DRAM designs are just another IP product to be bought and sold. An example of the new breed of IP company is Silicon Access, Inc. of San Jose. Comfortably set with major VC backing, it aims to make a business selling embedded DRAMs to fabless chip makers as well as to system-chip designers.

Independent purveyors of RAM cores differ from the old guard in that their embedded DRAM designs were created from scratch to be implemented with a logic process, thus promising lower cost and tighter integration. In contrast, embedded DRAMs from traditional DRAM vendors were created with DRAM processes that must be somehow converted to a logic process, commonly by adding large numbers of mask steps that will drive up the final chip cost.

Regardless of who wins, the newcomers or the old guard, this confluence of different technologies will create a whole new generation of complex systems on a chip: single-chip set-top boxes, cell phones, DVD players, and more. Dataquest agrees. It estimates that embedded DRAMs will represent a $4.4 billion market by the year 2000. Indeed, having DRAM cells will be a prerequisite for remaining competitive in the ASIC market. *

To voice an opinion on this or any Integrated System Design article, please e-mail your message to miker@isdmag.com.

integrated system design  February 1998