Roadmap of AI Processors

Ambient has developed the groundbreaking DigAn AI architecture that can scale from 5-cores to 8000 cores and from 40nm technology to 7nm or even 5nm technology. With such versatile and scalable architecture, Ambient Scientific team plans to build a roadmap of AI processors to address range of Edge AI applications starting from Endpoint AI devices. Our first products in the family of GPX AI processors, GPX-10 and GPX-5 are here.

ai processor

Distinctive Features:

  • Ambient AI processors are built using DigAn Architecture. DigAn uses revolutionary Cubic computing Concepts to achieve high performance with Ultra-Low-Power consumption
  • Each core is Software defined and programmable
  • Support for multiple operand resolutions for both weights and inputs. Operand resolution can be set and modified dynamically by the application to optimize performance and power consumption
  • Each core can be enabled, disabled and halted dynamically
  • Character Morphing between Inference and training – GPX processors can switch modes between training and inference dynamically under application control
  • Modern and intuitive SDK makes it easy to develop and optimize custom AI algorithms.

GPX-10 AI Processor

GPX-10 is a high performance, ultra-low power AI processor for Always-On embedded AI applications using deep learning algorithms. GPX-10 is a single chip AI computer. It is an integrated SoC, with 10-DigAn AI cores, 8 simultaneous sensor inputs, ultra-low-power ADC, custom memories, ARM M4F processor and numerous peripherals

Ambient Scientific’s full featured software development kit (SDK) supports AI-algorithm and model development using Tensor flow framework with Pytorch, Cafe, and other popular AI frameworks coming in the near future.

Ultra-Low Power AI for Always-On applications
  • Power consumption of 80 micro watt or less for Always-On wake word or anomaly detection
  • Runs Always-On AI applications for multiple years on a single coin-cell battery
  • Charger-less AI applications possible by harvesting energy from various natural sources such as kinetic energy and RF energy
Highly Integrated
  • Multi-channel ADC
  • Sensor fusion – Connect up to 8 analog and digital sensors simultaneously
  • ARM M4F for executing non-AI workload
Character Morphing between Inference and re-training
  • Collect and label training data at endpoint without connecting to the internet and cloud
  • Executes both inference and re- training applications. Switch from one to other mode dynamically under application control
Versatile and Flexible
  • Software-defined Programmable AI cores
  • Support for multiple operand resolutions for both weights and inputs. Operand resolution can be set and modified dynamically by the application to optimize performance and power consumption
  • Each core can be enabled, disabled and halted dynamically
Interchangeable Master-Slave Relationship
  • DigAn AI engine can change behavior between being a master or slave. This capability enables highly optimized, high performance and ultra-low- power AI applications
GPX-5 AI Processor

GPX-5 AI Processor

The GPX-5 is an AI inference processor. It is a smaller version of GPX-10 with only 5-cores. It can be used as a stand-alone SoC or as companion co-processor for AI inference. GPX-5 is an integrated SoC, with 5-DigAn AI cores, 8 simultaneous sensor inputs, ultra-low-power ADC, custom memories, ARM M0 processor and numerous peripherals.

Key Features:
  • 5 AI cores provide 1028 MAC operations per clock cycle
  • ~60 uW active power consumption in inference tasks
  • Sensor fusion with 8 I/Os
  • Years of operations on a coin cell battery
  • Lower cost than the GPX-10

DEVELOPER TOOLS FOR GPX PROCESSORS

Ambient SDK

Ambient Scientific provides tools for application developers to implement any neural network model and other non-AI functions onto GPX processors. The Ambient SDK lets you build your AI applications in popular languages then compile your code to your GPX processors. Build your own models and scale your IP across your GPX enabled devices with the Ambient SDK.

The Ambient SDK currently offers support for...
  • TensorFlow

...and will add support for

  • PyTorch
  • Caffe
  • Theanos
SDK1.jpg
Training Vault

Ambient Training Vault

Harvest your data from the GPX-10 and program on-device training with the Ambient Training Vault. Under a subscription model, you can retrieve your device collected data and use it to personalize applications, improve application performance or scale your AI-enabled ecosystem. Differentiate yourself from your competitors and build new business models for your products with the Ambient Training Vault.

Key Features:
  • Label data as it is captured
  • Harvest your labeled data from the GPX-10 via the Training Vault APIs
  • Own your application's data and use it in other AI applications

DigAn