Home > Steve Chen

Steve Chen, PhD'75, is a pioneer in the design and development of large-scale, parallel supercomputing systems. From Burroughs to Cray Research to his own companies, SSI and Chen Systems, Chen is recognized the world over as an industry leader and visionary.

Steve Chen was the principal architect of the most influential high-performance computer ever made: the Cray X-MP. Renowned as a supercomputing giant, it’s no surprise to many of us that Chen earned his PhD in one of the cradles of supercomputing might, the University of Illinois.

Chen spent most of his life in Taiwan, where he earned a BS in in electrical engineering from National Taiwan University in 1966. After graduation, he remained as a teaching assistant and spent a great deal of time with the school’s mainframes computers, an IBM 350/320, Control Data Cyber series machine, and IBM 1620. He spent hours as a teaching assistant helping students debug their programs, which were written in assembler. "Learn as we go. Teach as we go. Debug as we learn," recalled Chen. By interpreting the binary code, he started to learn how the machine’s logic worked. "I really got to understand how computers worked in a basic way," he said, "before learning the higher level applications, like FORTRAN and COBOL, and how they were compiled. I liked the computers. Then I learned how to use them. Then I wanted to understand them."

Chen left Taiwan to attend graduate school at Villanova University in Philadelphia. There bought computer books and maintenance manuals from an IBM sales office and reverse-engineered a software model of a computer. Chen spent several months, with the help of his wife, typing up his software on punch cards—ten boxes of them! He found he could insert the cards into either his own emulator or into the actual machine he was emulating and come up with the same result. Then he followed his desire to design a real machine.

After earning his MS in 1972, Chen came to Illinois to work with Professor Dave Kuck and graduate student Duncan Lawrie, who were championing the new concept of parallelism in the ILLIAC IV project. The ILLIAC IV was an array processor that contained four quadrants of 64 processing elements each. The quadrants could be configured to work synchronously under a single control unit or as four independent processors.

"In those days," recalled Professor Duncan Lawrie, "we were mostly thinking about things like compiler generators and languages with semantics for parallel processing. Chen was one of the first to think about automatic detection of parallelism so that users wouldn’t have to worry about thinking parallel. It turned out to be a lot harder than any of us thought, but he made some very significant contributions to the field. And of course, along the way, he evidently learned a few things about high-performance architecture."

Professor Kuck assigned Chen a project to explore software techniques that would enable FORTRAN programs to run using this parallel concept. The supercomputer industry was in its infancy, and several Illinois alums were working in the area of uniprocessor supercomputers, like Ken Miura, MS’72, PhD’73, at Fujitsu and Toshiro Kunihiro, MS’59, at NEC.

"I believe Chen was the programmer of the first version of Parafrase, Kuck’s compiler system," recalled Professor David Padua, "and he contributed with many interesting compiler ideas. For example, I think it was he who developed the techniques we use today for loop interchanging."

When Chen graduated, he set out to develop a commercial parallel system. "I was very focused," he recalled. "I never thought of anything else. I joined Burroughs Corporation in 1975, got a prototype built, developed a vector compiler and hardware for the control unit, and got my feet in the water for real development." Chen’s prototype was the Burroughs Scientific Processor (BSP), a supercomputer with an array architecture that could compete with the newly developed Cray-1. But the finished product never got out the door. Chen left Burroughs in 1978 to develop a high-speed scientific array processor for Floating Point Systems, Inc.

After a year at Floating Point Systems, Chen joined Cray Research as its chief designer, where he led the development of the world’s most commercially successful parallel vector supercomputers, the Cray X-MP, and its successor the Cray Y-MP. Chen began by making some architectural changes to the Cray-1, which was introduced in 1971. In the Cray X-MP (Chen said that the "X" stood for "extraordinary"), Chen introduced shared-memory multiprocessing to vector supercomputing. The machine contained two pipelined processors compatible with the Cray-1 and shared memory. The X-MP series was expanded to include 1- and 4-processor machines. The X-MP4 was the first supercomputer installed at the National Center for Supercomputing Applications (NCSA) at Illinois (summer 1985).

The first of the Y-MP series, Cray’s new multiprocessor vector supercomputer introduced in 1988, contained 1 processor, followed by 8, and then 16. All these machines shared essentially the same architecture, and the majority were designed by Chen and his team. Cray Research enjoyed tremendous growth from 1982–86 as its customer base expanded beyond government laboratories to commercial applications. This was the "heroic age" of the supercomputing industry. Sales blossomed from $50 million to $500 million as other U.S. supercomputing companies like Thinking Machines, Kendall Square Research, nCUBE, and MasPar proliferated. It wasn’t until the personal computer revolution in the 1980s that the commercial supercomputing industry began to fade.

Chen spun off from Cray to start Supercomputer Systems, Inc. (SSI). This was a mutually agreeable arrangement for the two companies; it was simply too much for one company to support both the Cray series (now at Cray-3) and the Y-MP series. Partnered with IBM, SSI drew in some large corporate customers, like Boeing, Du Pont, and Ford. SSI built the world’s most powerful, highly parallel supercomputer of its time, the SS-1.

In the 1990s, the supercomputing market had completely changed. "The cold war was over," Chen explained. "The commercial world was so competitive, and companies couldn’t afford the machines. Meanwhile, microprocessors became more powerful. It was the beginning of commodity technology instead of proprietary technology. There was the move to vertically integrate everything. We were in a new age. Supercomputers were too expensive. They were not only hard to build, but they took a long time to build."

"Everything we used in the 1980s," Chen said, "we pushed to the extreme limit. It’s very expensive to sustain this. You can only have so much volume with microprocessors, so we are building alternatives. In general, a lot of the ideas are similar, but they are applied in different places. For example, before, every five years you had to redesign the floating point micro division. Now the building blocks are bigger. You use processors instead of gates."

In 1993, Chen founded SuperComputer International (SCI) which was later renamed Chen Systems. The company developed Intel-based, 8-processor mid-range open enterprise servers (CS-1000) based on commodity component technologies. In 1996, the company was acquired by Sequent Computer Systems. Chen was most recently Sequent’s Executive Vice President and Chief Technical Officer. He left the company this summer, and we await to see what he does next.

Like many others in the field, Chen believes that future supercomputers, if they are to be commercially viable, must be based on Intel, commodity-based building blocks. They must be able to tackle both business and scientific problems so that multiple users can use the same machine. Therefore, the machine must be sharable, partitionable, integratable, as well as cost effective and highly available. Chen predicts that these machines will revolutionize the IT and platform industries. "The next wave will be in system technology," he said. "The endgame is commoditized, supercomputing knowledge—making it available to businesses and communities for things like videoconferencing, entertainment, shopping, and so on. You can see us merging into a new era of massive, deployable software."

In November of 2004, Chen visited Cornell University and gave a talk on the importance of compilers and optimization techniques to decrease size of executables. This is important for the smaller L1 caches found in most computer architectures.

Chen is a member of the National Academy of Engineering and the American Academy of Arts and Sciences. He lives with his wife and their two youngest children in Oregon. Their two older children are grown.