Mixed-signal chip enables fast low-energy wireless baseband processing
A 105 MS/s 4.18 nJ/inference 0.765 $\text{mm}^\text{2}$ Mixed-Signal Streaming ML Engine for Wireless Baseband in 12LP+ FinFET
Hardware Architecture
Summary
Processing signals for wireless communication usually requires a lot of power and space. The authors present a tiny chip called SIXPPAC that can quickly process wireless baseband data while using very little energy. It combines analog and digital parts to run a small neural network directly on the incoming data. This design can handle signals in real time and fits all its necessary parts on a very small piece of silicon. The chip’s energy efficiency and speed can help build smarter and more efficient wireless devices.
What this means in practice
- •For wireless hardware engineers: Integrate a low-power chip that performs real-time machine learning on wireless baseband signals to improve sensing and communication capabilities.$Commercial implications: Enables compact, energy-efficient wireless devices with machine learning for integrated sensing and communication functions.
- •For internet of things developers: Use a mixed-signal processing engine to add real-time sensing and communication intelligence to small IoT devices with strict power budgets.
Authors
Oscar Castañeda, Lavinia Recchioni, Sueda Taner, Thomas Burger, Jérémy Guichemerre, Christoph Studer
Abstract
We present SIXPPAC, the first streaming machine-learning engine for real-time integrated sensing and communication (ISAC) in wireless systems. SIXPPAC integrates two SAR ADCs and a digital binary-weight six-layer neural network (NN). The 0.765 mm$^\text{2}$ 12 nm FinFET engine stores all NN parameters on chip, enabling sustained 105 MS/s processing of 8b I/Q baseband samples at 0.8 V. By deploying bipolar inner products using AND instead of XNOR gates, SIXPPAC consumes 4.18 nJ/inference (0.42 POPS/W) at 0.8 V. We evaluate SIXPPAC on three example applications to demonstrate its versatility.