Thermal Management Techniques for Field Programmable Gate Arrays

摘要

Thermal aware design methodologies are gaining increased attention in the age of high performance computing. The emergence of three dimensional integration has opened up a world of possibilities in the realm of high density Field Programmable Gate Array (FPGA) based designs. Applications such as concurrency mining, deep learning and data mining are all being implemented on FPGAs due to the amazing flexibility and the performance per watt advantage FPGAs have over Application Specific Integrated Circuit (ASIC) based solutions. These types of compute-intensive applications bring with them the problem of efficient thermal management. Increased power densities lead to thermal hot spots and drop in performance. In this dissertation, we propose a placement level technique to minimize the thermal gradient of FPGA based designs. In the second part of the dissertation, we focus on the integration of microfluidics with FPGA designs. Microfluidic cooling involves the application of microchannel heatsinks for more effective cooling of integrated circuits. This technology is currently not in widespread commercial use due to technological and cost considerations but microchannel heatsinks are expected to become mainstream in the coming years. In this work, the performance of microhannel heat sinks is compared against air cooled heat sinks to demonstrate why air cooled heat sinks will not be sufficient to cool 3D FPGAs. The microchannel simulations were performed using a state of the art academic thermal emulator to determine optimal microchannel designs. A flow to mitigate the impact of coolant self heating is also proposed.