We are passionate about pushing the envelope and bringing new technologies to market. Learn more about our technologies below.

Asynchronous Resilience

a fundamental breakthrough in energy-efficient computing


Asynchronous Resilience is REM's patented circuit technology that enables our chips to operate drastically more efficiently than competitors. Asynchronous Resilience builds off of decades of research and combines:

  • Asynchronous Computing - 99% of processors are synchronous, meaning each operation is allotted the same amount of time. Asynchronous processors allows simpler operations to finish sooner, saving time and energy while requiring no change in software.
  • Timing Resilience - Timing resilience allows a processor to automatically adapt to changes that may occur in timing due to process, voltage, and temperature (PVT) variations.

This combination offers new flexibility and enables a single core is able to operate optimally across a wide power/performance range. When compared to synchronous solutions, REM circuits offer higher peak performance, greater efficiency at all voltages, and are able to operate at near-threshold voltages where transistors are fundamentally more energy-efficient.

REM is the only company to successfully implement timing resilience, and all known synchronous attempts have significant issues. We develop asynchronous designs with completely standard tools, cells, and processes.


Industry-Leading Efficiency

REM designs offer greater efficiency when operating at the same performance as synchronous solutions. The timing-resilience allows REM circuits to operate at near-threshold voltages where transistors are fundamentally more efficient.

Dynamic Power / Performance Tuning

The self-timing and resilient nature of REM's circuits allows a single design to operate across a wide power/performance range, and to move across this range within nanoseconds. Power when you need it, savings when you don't.

Universally Applicable Technology

Unlike other power-saving optimizations, REM's asynchronous resiliency is universally applicable to any logic block. REM's internal tools allow automatic conversion of synchronous IP, enabling REM to quickly improve on existing designs.
Request Whitepaper

AI Core

a radical new architecture for accelerating neural network inference


Deep learning has quickly become the most powerful tool for artificial intelligence and computer vision problems. Despite the popularity and promise of neural networks, their computational complexity has made them infeasible for many devices.

AI Core is a completely new hardware architecture built to accelerate local neural network inference and bring AI to new classes of devices. It's been designed to offer both superior performance and energy-efficiency by intelligently managing a massively-parallel array of neural processing elements. AI Core is over 20 times more energy-efficient than competitors because it leverages our years of experience in low-power design, asynchronous network-on-chips, optimized math circuits, and hardware architecture.


Incredible Efficiency

AI Core offers a 20x leap in efficiency over today's best solutions. Every aspect of AI Core's architecture has been fine-tuned for maximum energy-efficiency, from minimizing data movement and computation, to efficiently parallelizing inference across the ultra-low power processing elements.

Blazing Performance

AI Core enables a new level of intelligence for edge devices, offering over 100 FPS on many popular neural networks. AI Core's architecture is carefully designed to increase performance of convolutional neural networks, enabling data-center performance in an embedded form factor.

Ready for your Network

AI Core's flexible architecture has been designed to support a future-proof range of neural networks. REM Development Tools compile and optimize Caffe and TensorFlow networks for AI Core without any loss in accuracy, allowing you to easily bring existing networks to new devices.
Request Whitepaper