English EN Japanese JA

Technology

Technology

Scalable Architecture
Our architecture is scalable and can produce Co-processors from 0.5 tops to 64+ Tops by increasing number of agents.
Architecture Innovation
Our architecture has many advanced architecture innovations such as layer fusion in H/W, intelligent memory access and on-chip memory optimizations that give higher utilization.
Learning Agent or Inference Agent
Each agent can be configured as learning agent or inference agents. Our architecture enables learning at the edge within constrained resources.
Best-in-class Performance
Based on a novel hardware that utilizes small tensor units and a novel instruction set architecture that minimizes overhead and increases efficiency, AlphaICs solution provides best-in-class performance.
One-click Solution
AlphaICs software stack (AlphaRT) provides seamless environment for deploying neural networks onto the RAPTM. AlphaRT supports TensorFlow, and we plan to add support for other AI frameworks in the future.
Edge Learning
Edge Learning enables learning new features on the Edge. AlphaICs RAPTM architecture enables learning as new data arrives at the edge device without impacting previously learned intelligence to solve these problems.
Neural Network
A trained traditional neural network does not learn continuously while inferring. As the devices encounter new, previously unseen data, the system cannot adapt to the new input, and hence accuracy suffers significantly. We learned from many modalities asynchronously and regulated learning rate based on agent-based design.
Advantages
  • Less training data requirement
  • Continuous learning on-device
  • No need for retraining