Compilation time advantage of parallel-operation-oriented optically reconfigurable gate arrays

摘要

Recently, studies have matured of field programmable gate arrays (FPGAs) that realize hardware acceleration. For such hardware acceleration on FPGAs, hugely parallel computation is frequently used. Consequently, numerous identical circuits are implemented onto an FPGA. However, identical configuration contexts of numerous identical circuits are stored on different regions of the configuration memory when a using currently available FPGA, which is exceedingly wasteful. Therefore, we propose a parallel-operation-oriented ORGA with many gate array layers sharing a common configuration context. In programmable devices, the configuration memory and configuration circuit occupy a large implementation area: 70 %-80 % of the entire VLSI chip area. For that reason, the gate density of the programmable device could be increased if the amount of configuration memory were decreased. Therefore, a parallel-operation-oriented ORGA presents advantages in terms of gate density. Furthermore, the parallel-operation-oriented ORGA architecture presents the important advantage of shorter compilation time. This study clarifies the benefits of the parallel-operation-oriented ORGA in comparison to those of an FPGA having equivalent gate array structure based on the same process technology.