Zhimin Chen, Ken Eguro, Alessandro Forin, Ruirui Gu, Zhanpeng Jin, Paul Larson, Wenchao Li, Weiqin Ma, Rene Müller, Neil Pittman, William Bengston, Meg Davis, Drew Fisher, Larry Laugesen, Steve Liu, Grant Marvin, Jon Moeller, Brandon Nance, William Somers, and Jillian Weise
The Embedded and Reconfigurable Systems Group at Microsoft Research has been engaged in joint academic collaborations spanning both teaching and research activities. The results of these collaborations are showcased annually at the Faculty Summit at Microsoft headquarters in Redmond, WA, during the DemoFest event. This is also a good opportunity to review some of the other research projects that the group is engaged in, with special consideration to the research performed as part of the graduate internships. This report presents the demonstrations that took place during the 2009 DemoFest. We presented two undergraduate student projects from the Real Time Distributed System group at Texas A&M. One is a touch screen prototype that uses light diffraction rather than pressure-sensing to realize a multi-touch 2D input device. The second is a LED-input based dance pad that overcomes the wear and tear problems of traditional dance pads by using light sensing, and connecting LEDs as input rather than output devices.
Members of the ERSG group presented a number of research projects and demos. A set of APIs simplifies the communication between PCs and FPGA boards and when using Gigabit Ethernet achieves full-bandwidth speed. FPGAs are also used to accelerate the processing of networking protocols in a database system, with automatic generation of the circuits directly from the protocol specification. A new CPU model for the Giano full-system simulator supports the x86 ISA, and additionally realizes mixed concrete and symbolic execution of binary codes to detect data races in multi-threaded programs. A novel system for mining specifications deduces timing constraints in timed traces for digital circuits, embedded software, and network protocols. The system accurately pinpoints the source of errors in a faulty eMIPS micro-processor design. IEEE compliant Floating-Point execution units are fully optimized on a per-application basis and dynamically un/loaded in the reconfigurable logic portion of the micro-processor. The M2V compiler can now handle multi-basic blocks of MIPS binary code to automatically generate application accelerator circuits. A dual-core version of the eMIPS system demonstrated near-perfect speedup on the Montgomery modulo multiplication of large integers. The NetBSD Operating System runs in multi-user mode on the eMIPS system on two FPGA platforms. It uses a new, online scheduler to allocate the available accelerators slots to competing software applications.