Brief Description:

This design utilizes recent machine learning modeling based on Linear delay lines incorporates linear temporal dynamics for neurally-based classifier approaches. The compiled SoC FPAA classifier required 20-30μW for a range of classification tasks [1,3-5]. We expect this new custom classifier component to require 20-40μW when the μP is turned off. The additional general linear dynamics block which allows for the detection of patterns over time. This block effectively allows each output category to have a preferred temporal signal instead of being classified based on a single spectral pattern.

This effort looks to compile this acoustic classifier network in a single Caraval Skywater platform utilizing and modifying the analog Floating-Gate (FG) standard cell library to place and route the final structure. The analog FG standard cell library, based upon experiences in FPAA devices, is currently in fabrication through Skywater 130nm CMOS.

If you are collaborating on this project, please click here to access your collaboration files, and click "Accept Share" in the actions column if you haven't done so already.

project layout image
project layout image
Layout Image

Praveen raj


This Skywater 130nm implementation project will build a programmable end-to-end microphone-to-symbol keyword classifier. This full system classifier extends the large-scale Field Programmable Analog Array (FPAA) command-word classifier to allow for larger vocabularies (100-500) by accounting for temporal dynamics in the classification.




Floating gates

Keyword Spotting

low power


Last MPW Precheck


12/22/21 19:00:55 PST