| Field Programmable Gate Arrays (FPGA) are being used to develop a new hardware simulation technique that can provide a million-times speed increase for researchers modeling thermal phenomena coupled with other multi-physical processes.
“Multi-physics” systems involve the interaction of different processes, including thermal, chemical, mechanical, and electrical processes, and are found in activities ranging from bio-technology to controlling air bags. System design is complicated by the many interactions, and it is particularly difficult to predict multi-physics failure modes. A system designed to operate primarily in one physical domain can be rendered inoperable by a phenomenon in another domain. For example, a MEMS accelerometer in an airbag control system could exhibit a significant frequency shift due to heating caused by digital logic power dissipation in the area near the accelerometer.
Historically, multi-physics system design has involved expensive laboratory-based environmental testing, or through even more expensive testing in the operational environment. Moving to a simulation-based design environment has been restricted by the excessively long computation time of software-only modeling of these complex systems.
The new methodology being developed is based on the cellular automata approach that requires solutions to numerous first order equations in series, and FPGA simulation techniques. “FPGAs provide the ideal mechanism for acceleration of these models,” said Jim Armstrong, the ECE leader on the project.
This is because FPGA cells exchange information from the previous iteration with their nearest neighbors, he explained. “The solution rate is the cell clock rate. A conservative value for an FPGA cell clock rate is 10 MHz. Thus, a case with 10,000 iterations would require 1 millisecond. The typical simulation time for the same case using Mathematica would take about 10 seconds.”
Implementing the models on FPGA will involve several challenges for the computer engineers, including issues in fixed point arithmetic, FPGA cell architecture and execution modes, and run set-up and management. “The history of computers has demonstrated that significant speedup of computation by tightly coupled parallel processes is only achieved by specialized hardware for particular types of algorithms,” Armstrong said. “Our work will contribute in this area.”
|
