ABOUT
Analog-to-Feature Extraction
Task 2.1
Hybrid Algorithms for Feature Extraction
Task Lead: Justin Romberg, Other PIs: Stanley Chan
OBJECTIVE
This task will explore the computational building blocks for informative dimensionality reduction to reduce the data both before and after digitization, allowing the sensor to provide relevant information for the task at hand while making efficient use of available resources.
Task 2.2
Analog Feature Extraction (AFE)
Task Lead: Michael Flynn, Other PIs: Saibal Mukhopadhyay
OBJECTIVE
This task will design innovative mixed-signal circuits for extracting spatial, temporal, and spatiotemporal features directly from the analog signals to reduce data dimension. Deliverable. Design and test-chip measurement results of AFE cicuit blocks.
Task 2.3
Multimode Passive Imaging
Task Lead: Hun-Seok Kim, Other PIs: David Blaauw
OBJECTIVE
The objective of this task is ti develop on hyper-spectral passive imagers monitoring blackbody radiation and solar reflection from the environment, at ultra-low power consumption. Deliverable Develop integrated passive imaging covering multi-spectra from mm-wave, THz to infra-red in heterogeneous integrated platform.
Task 2.4
Digital Feature Processing (DFP) Engine
Task Lead: David Blaauw, Other PIs: Hun-Seok Kim
OBJECTIVE
This task develops highly energy-efficient, versatile, and high performance neural network computing engines for the digital feature processing (DFP) platform that support real-time feature streams from multi-modality AFEs while at the same time operating at very low power. Deliverable. Design and measurement results of test-chip on DFP circuit blocks.
Task 2.5
Performance Limits and Information Analysis
Task Lead: Stanley Chan, Other PIs: Pamela Abshire and Justin Romberg
OBJECTIVE
The purpose of this tasklis to investigate the fundamental performance limit of the feature extraction algorithms developed in A2I and provide guidance to optimize system design to improve performance. Deliverable. A set of physically motivated models, benchmark evaluations, and recommendations for A2I system.