Accuracy of Digital, Efficiency of Analog
Meet the industry’s newest architecture for AI computing on-the-device and at-the-edge. DigAn™ by
Ambient Scientific is the AI architecture foundation of a new generation of AI edge processing.
DigAnTM is designed for AI
- 1000x lower power consumption (~8 µw per core) than ARM for inference
- 250 GOPS with 10 AI cores for more processing than X86
- Scalable with CMOS process for low-cost manufacturing
- Programmable accuracy: 4-bit to 32-bit for power optimization
On-device inference and training
- Architecture built for neural networks to run real-time inference
and on-board training
Integrated ADCs, Clock, and SRAM
- ntegrated MCUs, ADCs, and master clocking to reduce system complexity and total system costs
- Fuse analog sensor data from 8 onboard I/Os with 1 Mesasamples/second
- AI-SRAM with 5x less active power, 3x less leakage power
- On-board data sort and tagging for incremental learning
Analog Matrix Computing
The DigAnTM architecture is designed to accelerate neural network models for AI inference.
Each neuron is defined at the hardware level. This gives the DigAnTM architecture its speed
efficiency. DigAnTM enables our GPX processors to perform inference and training on the
device so models
aren’t stored in a data center, and inference doesn’t occur in the cloud. DigAnTM is on-device AI.