AI Accelerator (NPU) IP - 3.2 GOPS for Audio Applications

Overview

The Expedera TimbreAI T3 is an ultra-low power Artificial Intelligence (AI) Inference engine designed for noise reduction uses cases in power-constrained devices such as headsets. TimbreAI achieves a best in class <300uW power consumption due to Expedera’s unique packet-based architecture. TimbreAI requires no external memory access, saving system power while increasing performance and reducing chip size, processing latency and system BOM costs.

Designed to provide quick and seamless deployment into customer hardware, TimbreAI provides optimal performance within the strict power and area application requirements of today’s advanced audio devices. TimberAI’s unified compute pipeline architecture enables highly efficient hardware scheduling and advanced memory management to achieve unsurpassed end-to-end low-latency performance. This minimizes die area, saves power, and maximizes performance.

Key Features

3.2 GOPS; ideally suited for audio AI applications such as active noise reduction
Ultra-low <300uW power consumption (TSMC 22nm)
Low latency
Neural Networks supported include RNN, LSTM, GRU
Data type supported include INT8 x INT8, INT16 x INT8, INT16 x INT16
Use familiar open-source platforms like TVN, TFLITE, ONNX, TVM
Delivered as soft IP: portable to any process

Benefits

Ultra-low power implementation for battery-powered devices.
Drastically reduces memory requirements.
Portable to any process.
No DRAM required.
Run trained models unchanged without the need for hardware dependent optimizations.
Deterministic, real-time performance.
Improved performance for your workloads, while still running breadth of models.
Delivered as soft IP.
Simple software stack.
Achieve same accuracy your trained model.

Block Diagram

Applications

Battery-powered headset
Hearables

Deliverables

RTL or GDS
SDK (TVM-based)
Documentation

Technical Specifications

Maturity

In production

Availability