Accuracy and Scalability 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 foundational technology suite for a new generation of AI at the edge.

AI computing

Solving the three primary bottlenecks of on-device AI requires reinventing conventional computing architecture. Ambient Scientific built its DigAn architecture from the ground up. Taking the best aspects of digital and analog computing for AI applications, the DigAn architecture is optimized for inference and training with neural network models. You can create new business models, products, and scalable IP with the DigAn technology suite by Ambient Scientific.

AI Workloads

DigAn is Designed for AI Workloads

The DigAn architecture is designed specifically for AI applications. We've overcome the obstacles to creating scalable, low-cost, low-power AI hardware.

  • Low power: 10x lower power consumption than best-in-class industry solutions (~8 uW per core).
  • Scalable: Manufactured with CMOS process to scale from 40nm to 7nm and beyond.
  • High performance compute: 256 MAC per AI core per clock cycle @ 40nm node; scales to 2048 MAC @ 7nm node.
  • Programmable operand resolution: Developer can select 4-bit to 32-bit resolution to optimize power consumption for any AI model.
  • Dynamically change AI model resolution: Operand resolution can be selected in the application and changed at runtime.
DigAn Ambient's Analog Cuts AI Power White Paper GPX-10 product brief

Optimized for On-device Inference and Training

  • High MAC count: The DigAn matrix compute-engine in our AI core is optimized to quickly execute MAC operations in neural networks.
  • On-device weights and activation functions: All neural network parameters are software-defined and programmable.
  • Real-time AI: DigAn architecture enables on-device inference and training in real time.
  • Character Morphing: Dynamically switch between inference and training based on application requirements.
On-Device Inference and Training
Integrated Architecture

Integrated Architecture

  • Highly integrated: The DigAn architecture integrates an ARM M4F, multi-channel ADC, oscillator, and more to reduce system complexity and total BOM cost.
  • Sensor fusion: Use onboard I/Os to acquire data from multiple analog or digital sensors and sensor elements simultaneously.
  • Always-ON AI pipeline: Ingest sensor data for wake-on-event and anomaly detection, even when the onboard processor is in power-down mode.

© Copyright Ambient Scientific Inc. , All rights Reserved.