A Universal Maximum Likelihood GRAND Decoder in 40nm CMOS
Arslan Riaz, Muriel Médard, Ken R. Duffy, Rabia Tugce Yazicigil · 2022
As optimally accurate Maximum-likelihood (ML) decoding is NP-complete in code-length, all existing decoders are code-specific and so restricted in code-rate and code-length. Guessing Random Additive Noise Decoding (GRAND) is a recently introduced theoretically universal decoder suitable for all moderate redundancy codes, as required for ultra-reliable low-latency communication by the 5G New Radio Standard. Here we demonstrate the first-integrated universal GRAND chip in 40nm CMOS technology. It efficiently decodes all binary linear codes of any rate from 0.66 to 1 with code length up to 128 bits. For a bit-flip probability of 10−5, GRAND chip measurements show that it achieves 1 μs latency at 68 MHz. Dynamic clock gating leveraging noise statistics reduces the average power dissipation to 3.75 mW at 1.1 V consuming 30.6 pJ/decoded bit of energy with a throughput of 122.6 Mb/s. The chip allows seamless swapping between codebooks with no downtime, enabling use by multiple applications simultaneously.