When I arrived at MIT in 1993, Anant was in the midst of his first startup, decidedly bitten by the entrepreneurship bug. The project I was on was Alewife, which Anant discusses below. Many of the ideas and breakthroughs in Tilera date back to the research we did during Alewife. At the time, I was young and naive, and did not understand yet how these advances might be brought to market.
Throughout history, often technology and innovation has preceded the market by decades. The best example of this that I can think of is the Digital Camera, which was invented in the mid-seventies, but took until the late nineties / early two thousands to really gain market foothold. Similarly, concepts of Tilera have evolved over the last twenty years of Anant’s career.
SM: Was MIT doing the types of startup activities when you started, that you have been doing later on? AA: Yes, MIT has always been doing that. In 1994, I did a startup called Virtual Machine Works where we did a logic emulation system containing several hundred FPGAs. That technology was acquired by IKOS, which in turn was acquired by Mentor Graphics. They are still shipping logic emulation systems today based on the same product we created in 1994. That was a reasonable business success.
In the late 80’s, early 90’s we also started a multiprocessor effort called Alewife. The idea was to put a bunch of processors together in a mesh network and support them with cache coherence and make them a parallel computer, which we did. We demonstrated that in 1994 as well.
SM: What was the commercial situation around Alewife at the time? AA: Around that time, we were working on parallel processing involving multiple processors and getting them to work together. It was very early and the sequential processor juggernaut performance was still being improved year after year. There was no commercial market for parallel processors yet.
A few folks from my group and I went out and did Virtual Machine Works which was a technology we had created to verify the chip that we built for Alewife. We didn’t feel that building another super computer was a wise business decision. That is why we did not build a company based on Alewife. However, a lot of the ideas – cache coherence, mesh technologies – that we came up with those in the early 1990’s, are now making their way into commercial chips, including our own, 15 years later. A lot of what we were doing then was really decades ahead of commercial impact.