Issue



System-On-a-Chip changes everything


06/01/1998







Robert Haavind

Editor in Chief, [email protected]

System-on-a-chip changes everything

With system-on-a-chip (SOC) technology (System LSI in Japan), the entire electronics industry appears to be on the verge of a radical transformation. Chips with 0.35-micron linewidths are becoming common, 0.25 is moving into the marketplace, and 0.22 micron is not far behind (reaching 0.18 is proving more troublesome). The densities possible at these linewidths allow a wide range of circuitry to be arrayed on the same die, and doing so can greatly reduce total systems cost while simultaneously enhancing performance: the proverbial more for less!

National Semiconductor (Cyrix) is already offering an SOC that incorporates all the circuits needed for a low-level personal computer at a very low price. Microprocessor chips have been coming out with an assortment of associated peripherals on them - bus, codec, graphics processor, cache memory, modem - cutting the costs and design effort needed to incorporate them into total systems. Game vendors (who oddly enough have become technology leaders because of their need to push capabilities for dynamic graphics) have for some time surrounded the ASICs that personalize their game stations with multiple cores on the same chips. Now they are moving to powerful processors, such as Hitachi`s SH-4, which put digital signal processing functions right into the computational flow. Wireless phones, which already combine multiple cores to achieve compactness and low-power operation, continue to add such additional functions as internet access and e-mail, all on a chip or two.

This is putting tough new requirements on wafer fabs. For one thing, there used to be a sharp delineation between DRAM and logic fabs, but system chips are beginning to incorporate not only SRAM but some DRAM as well. Some foundries are offering libraries of hard core functions, tailored to their specific design rules, that can be easily tacked onto a customer`s ASIC. This saves the design time required to fit a soft core to a specific fab`s requirements. But at the same time, it limits the designers to an all-purpose solution, when better cores may be available elsewhere. While cores designed within the same large electronics firm may all meet the design rules for that company`s fabs, the different EDA software used to design the circuits may make interfacing the cores and doing verification difficult or impossible. The Virtual Socket Interface Alliance (VSIA) is trying to come up with standards that will make interchangeable reuse convenient, but it remains to be seen whether this effort can succeed or whether narrow proprietary interests will dilute the effort.

Because SOCs make great sense, both technically and economically, they will inevitably move strongly into the marketplace. But selling libraries of software is so different from buying handfuls of parts from a distributor and wiring them together on circuit boards that their emergence will require a wide range of new business models. Semiconductor fabs will be at the heart of the revolution in the electronics business. It will take astute management to operate effectively in the new environment. Fabs will have to remain alert to customer requirements, and to opportunities to offer better ways of working with chip designers, whether at fabless firms or within internal design groups.

While rapidly changing process requirements can easily absorb all the attention of those responsible for operating fabs, it will also be necessary to monitor the evolution of SOC technology and to organize operations to meet customers` changing needs as well. A double challenge lies ahead.