We enable and enrich sensors with hidden intelligence for energy-starved smart devices.


Up to 500x more energy-efficient than existing solutions.

Complete solutioning

System-level solutions incl. neural network models optimized for the application.


Support for a wide variety of sensory data and neural network models.

Neural network acceleration for the edge

Breakthrough energy-efficiency

Our chips intelligently reconfigure themselves in real-time to consume the least possible amount of energy for the active task. delivering between 4 to 100 TOp/s/W of energy-efficiency.

Designed for smart sensing

  • Consumer devices such as hearing aids, wearables, AR/VR, and health monitoring.

  • Aerospace & automotive such as drones, condition monitoring and in-car intelligence.

  • Smart sensors for industrial IoT and other energy-starved applications.

Designed for easy development

  • Train your networks with Tensorflow or PyTorch.

  • Developers require no specialized understanding of hardware

  • Our Python-based optimization libraries maximize energy-efficiency and performance

Our core technology: Adaptiva

Adaptiva is a neural network accelerator designed to optimize its internal configuration to process the active task in the most energy-efficient manner possible. Depending on the workload, Adaptiva delivers between 4 to 100 TOp/s/W of energy efficiency. An example of dynamic workloads:

  • Always-on tasks

    • Continuous classification of temporal signals from a biomedical sensor

    • Motion detection and classification from accelerometers

    • Audio key phrase detection from a sound source

  • Interrupt-driven computationally intensive tasks

    • Image segmentation

    • Speech translation

    • Object detection and identification

Evaluation board: AdaptiveStorm

We are working with CSEM on building an evaluation SoC for Adaptiva, called AdaptiveStorm. A high level overview of the planned chip is as follows:

  • Dual-core RISC-V ASIC

    • Watchdog mode for ultra-low power

    • Nullhop mode for high throughput

  • Neural network co-processor: Adaptiva

  • Supported neural networks

    • Convolutional neural networks including resNet, VGG, MobileNet

    • Recurrent neural networks with multiple layers


  • Access to software emulation of Synthara’s accelerators and customization features from Q4 2021

    • Support available on request

  • Early access to AdaptiveStorm evaluation board from Q4 2022

    • Price on request

  • For more information, please send an email to contact-us(at)synthara(dot)ai or complete the contact form below.

Dammstrasse 16, 6300 Zug (HQ),
Josefstrasse 219, 8005 Zurich

Social media