Krste Asanovic came to ICSI as a graduate student in 1989, not long after the institute’s founding. Within a few years, he was making groundbreaking advances in machine learning technology—decades ahead of today’s AI renaissance.
Machine learning algorithms are based on vast numbers of calculations, often requiring billions of operations to complete one task. The neural networks being developed in the early 1990s were simple compared to those used today, but even so, researchers recognized that these algorithms would quickly outstrip the capabilities of normal computers. Without specialized hardware, it just takes too long and requires too much energy to perform the enormous volume of computation necessary for AI.
With funding from the Office of Naval Research, the Advanced Research Projects Agency, and the National Science Foundation, Asanovic and colleagues set out to solve this problem with a new processor architecture combining the flexibility of a general-purpose RISC processor with the ability to efficiently carry out compute-intensive tasks through vector execution. By 1995, they designed and successfully fabricated a programmable chip optimized for neural network training, known as T0. The key advantage of the chip’s Torrent architecture stemmed from incorporating both general and machine-learning-specific vector processing elements on a single silicon die.
The innovation represented an important step forward for AI, laying the groundwork for researchers to experiment with ever more complex neural networks. Within months, researchers at ICSI and across the U.S. and Europe were using the chip to accelerate work in sound synthesis, image processing, speech decoding and more. Today, a robust industry has grown up around the demand for ever more powerful hardware to support machine learning—a vastly expanded ecosystem compared with the uncharted frontier in which Asanovic’s team operated years ago.
As AI becomes increasingly integral to our daily lives—from driving a car to diagnosing disease—ICSI’s AI group continues to push the boundaries to advance tomorrow’s technology.
What made ICSI a good place to pursue these projects?
“This was a very big project, spanning neural network applications, software, and hardware. We worked across the whole stack, and the advantage of being at ICSI was that all those folks were there at the same place. It was a concentrated center to do this kind of research.
A big output of these research projects is the people. If you look at all the different folks involved in that project, a lot of them have been very successful since then, going on to places like Google and IBM. The papers and the specific technologies have an impact, but it’s really about the people and the impact they have, continuing on through the next generation, advising people, starting industries, and the broad infrastructure impact their work has had on everybody.”
Krste Asanovic
Former Senior Research Scientist; now Chief Architect at SiFive Inc.
This story was published in January 2026 as part of a retrospective series highlighting ICSI’s accomplishments and impacts over the years. To learn about our ongoing work, explore our Core Research Themes.