In 2016, Quadric's founders realized that the rapidly evolving landscape of edge AI was creating a fundamental problem for chip designers. Fixed-function neural processing units (NPUs) targeting yesterday's operators were becoming obsolete as new model architectures emerged. Multiple accelerator IPs, fragmented toolchains, and the inability to adapt to emerging AI workloads were pushing tape-out schedules further right and creating brittle silicon that couldn't keep pace with innovation.
So they built Quadric - a unified AI processor IP company with a breakthrough approach to on-device inference. The company's Chimera GPNPU (General Purpose Neural Processing Unit) architecture blends the machine learning performance characteristics of a neural processing accelerator with the full C++ programmability of a modern digital signal processor. This enables a single licensable processor IP to handle both neural network inference and traditional DSP/control code in one unified architecture, eliminating the need to partition applications across multiple processors.
Quadric's comprehensive software stack includes the Chimera SDK with a graph compiler that auto-compiles hundreds of models, supports ONNX import from PyTorch and TensorFlow, and provides extensibility through Python (ChiPy™) and C++ with an included LLVM compiler. The technology scales from 1 to 864 TOPS, delivering power efficiency optimized for single-batch latency, deterministic timing with instruction-encoded data movement, and one toolchain across entire chip lines. With customers spanning consumer electronics, enterprise, robotics, and automotive markets - where their solution supports ASIL B/D for safety-critical applications - Quadric is enabling the next generation of intelligent edge devices.
