My research at Sandbridge is based on my Ph.D. thesis and combines DSP, Java, and RISC execution. I was a co-founder the first year CEO and now run the engineering organization as CTO & EVP, Engineering. Unlike many methodologies that take small snippets of benchmarks and extrapolate results, at Sandbridge we have built a complete Software Defined baseband. The core technology is a multithreaded compilable ultra-low power Java DSP with SIMD/Vector capability exactly as my thesis. We have developed unique techniques to dramatically reduce the power.
In less than 2 years we have working software, silicon, boards, and complete communications systems including RF.
At IBM Research I continued to investigate compilable DSP architectures.I was Manager of Advanced DSP Technology for Research in Yorktown but also managed the RTP Broadband Transmission group for Microelectronics. I was also the Access Aggregation Business Development Manager for IBM Microelectronics. My groups made some good progress, particularly on the e-lite architecture. Eventually IBM decided not to go into the DSP business and licensed LSI's ZSP. Myself and some colleagues then formed Sandbridge Technologies.
I worked in the Wireless Communications group in the Microelectronics division at Lucent. This company is now called Agere.
I was the Chief Architect for next generation wireless DSPs. My group designed a scaleable architecture for wireless and multimedia applications. One of my main focus areas was designing DSP architectures that make good compiler targets.
My group became part of Starcore when Lucent and Motorola decided to jointly development new DSP architectures. While at Starcore I managed the software effort.
I used to work in the Digital Signal Processing group at IBM Microelectronics. My key interests there were parallel DSPs for real-time multimedia (especially video compression and 3D graphics).
This research focussed on the Mfast processor (Mwave Folded Array Signal Transform Processor). It is a radically new parallel VLIW DSP with beaucoup performance for most signal processing applications. Most of the original designers went to Billions of Operations Per Second after Mwave folded.
My particular interest in M.f.a.s.t. was System Architecture, Processor Architecture, and Parallel Algorithms. In the area of parallel algorithms I have found tensor algebra to be very useful in mapping signal processor algorithms to parallel architectures.
It was at IBM that I met Dr. Jerry Pechanek who introduced me to prof. dr. Stamatis
Vassiliadis. These two got me really interested in parallel computer
architecture and the rest is history!
My research was primarily directed towards parallel computing, architectures, and algorithms. Specifically, I designed the Delft-Java Processor. It is a parallel multithreaded machine optimized for the Java language but also containing DSP/SIMD instructions for multimedia execution. Since Java bytecodes are interpreted, the goal was to speed up the interpretation of the program to a performance level that meets or exceeds natively compiled code. We found that with architectural support for just a few Java bytecode, we could effectively speed up most of the Java processing. A good overview is here. We also created an architecturally independent method for accelerating class resolution. Rather than using so called "fast bytecodes" which require self-modifying code, we invented the Link Translation Buffer (LTB).
Additional research involved reconsidering fundamental computations and the interactions of compilers with architectures. This can lead to new ways of designing systems. One of the new techniques is Reconfigurable Computing. Using FPGA technolgy, it is now possible to dynamically reconfigure target "architectures" on-the-fly. The result of this is the ability to more precisely match the computation to an architecture. This opens up entirely new ways of thinking about parallel architectures and compilers.
. Meet my advisor prof. dr. Stamatis Vassiliadis. He runs the Computer Engineering Department at T.U. Delft.
Previously, I was a student at the University of North Carolina at Chapel Hill. I studied computer architecture, parallel architecture, parallel languages, computer graphics, and distributed computing - as well as studying Dutch in my spare time! I also worked on the pixelflow parallel image processing/graphics machine.
Dr. Fred Brooks was my UNC advisor
? Copyright1995-2018 by John Glossner.