Embedded Systems

A scalable risc-v hardware platform for intelligent sensor processing

by Paul Palom­ero Bernardo, Patrick Schmid, Oliver Bring­mann, Mo­hammed Iftekhar, Babak Sadiye, Wolf­gang Mueller, An­dreas Koch, Eyck Jentzsch, Axel Sauer, Ingo Feld­ner, and oth­ers
In 2024 De­sign, Au­toma­tion & Test in Eu­rope Con­fer­ence & Ex­hi­bi­tion (DATE), pages 1–5, 2024.

Ab­stract

This paper pre­sents a demon­stra­tor chip for an in­dus­trial audio event de­tec­tion ap­pli­ca­tion de­vel­oped as part of the Scale4Edge pro­ject. The pro­ject aims at en­abling a com­pre­hen­sive RISC-V based ecosys­tem to ef­fi­ciently as­sem­ble well-tai­lored edge de­vices. The chip is man­u­fac­tured in Glob­al­foundries’ 22FDX tech­nol­ogy and con­tains a RISC-V CPU with cus­tom In­struc­tion-Set-Ar­chi­tec­ture Ex­ten­sions (ISAX) for fast AI and DSP pro­cess­ing, a low power neural net­work ac­cel­er­a­tor, and a scal­able PLL to ful­fill real-time pro­cess­ing re­quire­ments. By au­to­mated in­te­gra­tion of these spe­cial­ized hard­ware com­po­nents, we achieve a speedup of ×2.15 while re­duc­ing the power by 27% com­pared to the unp[ntimized so­lu­tion.