A Sub-0.8-pJ/bit Universal Soft-Detection Decoder Using ORBGRAND
Arslan Riaz, Alperen Yasar, Furkan Ercan, Wei An, Jonathan Ngo, Kevin Galligan, Muriel Médard, Ken R. Duffy, Rabia Tugce Yazicigil
Abstract
Guessing random additive noise decoding (GRAND) has enabled the practical implementation of maximum likelihood (ML) or near-ML decoding, shifting the paradigm of code-specific decoder design to a code-agnostic decoding architecture. Ordered reliability bits GRAND (ORBGRAND) is a soft-detection variant of GRAND that uses soft information to guide its query order to significantly improve the decoding performance. This work presents the first-integrated energy-efficient hardware implementation of ORBGRAND to achieve ultra-low energy (sub-pJ/bit) and power consumption (5 mW) while using a small core area of 0.4 mm2 in 40-nm CMOS. The proposed architecture enables dynamic power savings by implementing an efficient sorter that allows the decoder to use the sorted bits immediately without waiting for the entire list to be sorted and an efficient landslide unit that generates noise effect sequences in parallel. The chip is implemented in 40-nm CMOS with a re-configurable architecture that enables decoding of any binary linear code from 32 to 256 bits of code length and 0.8–1 code rate. For a code length of 256 bits and a code rate of 0.94, it provides a measured energy consumption of 0.76 pJ/bit and power consumption of 4.9 mW from a 1.0-V supply voltage at an operating frequency of 90 MHz providing a throughput of 6.5 Gb/s and a latency of 40 ns at a targeted frame error rate (FER) of 10-7.