TimbreAI T3
Embedded AI for Audio
Even the smallest, lowest-power audio devices embed AI capabilities to enhance the user experience. Successful deployment of AI on resource-constrained products like headsets and wearables entails careful attention to power consumption and silicon area requirements.
Power-Sipping, Always-Sensing AI
The TimbreAI™ is an ultra-low-power AI Interface engine designed for audio noise reduction use cases in consumer devices such as wireless headsets. It provides optimal performance within strict power and area constraints. Featuring 3.2 billion operations per second (GOPS) performance, the TimbreAI T3 sips an astonishingly low 300µW or less power. TimbreAI supports quick and seamless deployments. It is available as soft IP and is portable to any foundry silicon process.
Innovative Architecture
The TimbreAI is purpose-built for audio noise reduction in power-constraint devices. It uses Expedera’s packet-based architecture, and use case optimizations to achieve impressive performance and power efficiency.
Run Your Trained Models Unchanged
The T3 requires no changes to your trained models and no sacrifices to accuracy or performance to achieve your desired PPA goals.
Pre-Configured for Audio Neural Networks
The T3 is pre-configured to support common audio neural networks.
Ultra-Low-Power AI Interface
Reducing power consumption to an absolute minimum is essential to product success; the T3 has been architected to minimize dark silicon and requires no external memory, consuming less than 300 μW of power.
Successfully Deployed in 10M Devices
Quality is key to any successful product. Origin IP has successfully deployed in over 10 million consumer devices, with designs in multiple leading-edge nodes.
Use Case
AI-enabled, Long Battery Life TWS Headsets
A chipmaker wanted to include AI-enabled audio denoising on its next TWS (True Wireless Stereo) headset products. However, even with a small performance load of 3 GOPS, the chipmaker found that off-the-shelf AI processing via MCUs or NPUs was too power-hungry for a satisfactory user experience. Using TimbreAI, the chipmaker reduced power consumption and AI chip area by 50%.
|
|
|
|
|
| Compute Capacity | 3.2 GOPS |
| Power Efficiency | 300 μW |
| Layer Support | Standard NN functions |
| Data types | INT4/INT8/INT16 Activations/Weights |
| Quantization | Channel-wise Quantization (TFLite Specification) |
| Latency | Deterministic performance guarantees, no back pressure |
| Frameworks | TensorFlow, TFlite, ONNX, others supported |
Download our White Papers
