The Next Big Innovation in Microprocessors: Anant Agarwal (Part 3)

SM: What came after Alewife and VMW? AA: I did VMW in 1994 – 1995, and in 1996 I came back to MIT. I started the Raw effort in 1996. Looking at processor design, we felt that in another 10 years we would have chips with billions of transistors and we wanted to discover how you would build processor chips in that era.

We came up with a basic multicore design, and we published that design in IEEE Computer in 1997. That was a design we published which had 16 tiles, and they used a mesh interconnect. We got funding from DARPA and began working on the project in 1996. We demonstrated the Raw project, which was a 16 core tiled multicore chip, in 2002. We collaborated with IBM in building that chip.

When we published the paper in 1997, the editors asked some research groups what processor chips of the future would look like.
Our design, multi-tiled / multi-cored, was how we felt chips should be built because of the problems which will be faced by trying to make single processor chips grow more and more powerful. There were some other designs in that issue as well. One design simply said processors would get bigger and bigger, another design talked about vector machines, so there were many different proposed designs. We seemed to have hit a good design with the tiled, multicore approach, which seems to match the direction the market is going in now. As I mentioned earlier, we demonstrated that chip in 2002.

SM: In 2002 you had the 16 core chip. AA: Yes, we had that chip finished and successfully demonstrated it.

SM: When did the idea of taking the tile technology out and building a company come together? AA: Interestingly, multicore technology was not hot at that time. Single processors seemed to still be continuing in the late 1990’s through 2001. I think 2002 was the year when single processors hit the brick wall. We were validating the ideas in the Raw chip, validating software systems and compilers, in 2003. Once we had the chip and the boards, we could experiment with it in 2004. I decided to form the company in mid-2004. One of my VC friends had been following the work in Raw, and he felt this was a good opportunity. I had been holding him off for the years.

[To Be Continued]

[Part 2]
[Part 1]

This segment is part 3 in the series : The Next Big Innovation in Microprocessors: Anant Agarwal
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15