Maximizing the throughput of threshold-protected AES-GCM implementations on FPGA

In this paper, we push the limits in maximizing the throughput of side-channel-protected AES-GCM implementations on an FPGA. We present a fully unrolled and pipelined architecture that uses a Boolean masking countermeasure (specifically, threshold implementation) for first-order DPA resistance. Using a high-end Virtex-7 device, we obtain a throughput of 15.24 Gbit/s. Since masked implementations require a stream of random bits for each execution, a high-throughput masked implementation requires a high-throughput pseudorandom number generator as well. This work determines how fast random numbers should be generated in order for ultra-high throughput, threshold-protected AES-GCM implementations to be feasible on FPGAs.