The originator of much of the technology underlying today's superscalar out-of-order microprocessors, Metaflow was never able to reap its just rewards. Begun in 1985, the company's first project, a SPARC-based ECL gate-array processor, was supplanted in 1989 by Lightning, a CMOS design backed by funding and IC-design resources from Hyundai. But the division of labor between the two companies proved stormy, and in 1991 Lightning was discharged, creating Thunder--a 0.8-micron three-chip processor designed entirely by Metaflow (see MPR 2/12/92, p. 9). In 1995, the company successfully demonstrated an 80-MHz Thunder processor--which delivered 2.5 SPECint92/MHz--just as Hyundai decided it wanted x86-based processors instead of SPARCs.
Metaflow had a brush with success in the early '90s when, working under contract to Intel, it convinced the processor giant that out-of-order design offered the best hope for building fast x86 processors. In fact, sources indicate that inside Intel the P6 was initially referred to as the "Metaflow processor." Sources have also revealed that Intel actually tried to buy Metaflow, but Andy Grove, unable to come to terms with Hyundai, had to abort the purchase.
Intel subsequently acquired the IP rights to Metaflow's technology through a patent cross-license with Hyundai for DRAM technology. Today the P6, with its centralized reservation station and instant repair of out-of-order and speculative execution, bears the clear mark of Metaflow�s involvement. More than 150 Intel patents cite Metaflow intellectual property as prior art. In 1997, frustrated by the inability to control its own IP and thus its own destiny, Metaflow convinced Hyundai to sell out to ST. Since then, ST and Metaflow have been jointly developing an x86-based processor. But, already unhappy with ST's plans for that part as well as with its lackluster process technology, Metaflow's leaders were pushed over the brink when ST traded Metaflow's IP to IBM in an x86-for-PowerPC swap (see MPR 8/3/98, p. 10).
Some Metaflow designers continue to work on ST's x86 project, but with the leaders gone and ST more interested in the low end of the x86 market, chances are slim that the x86 chip will ever see daylight--especially now that ST has a Pentium-class x86 project under way in India. ST officials declined to comment, other than to acknowledge that ST has indeed taken full ownership of Metaflow and to express confidence in the design team that remains.--K.D.
Copyright ©1999 by Microprocessor Report (April 1999 issue)