quadric, Inc
quadric develops and licenses a unified on-device AI processor IP architecture, the Chimera GPNPU, combining neural network inference with C++-programmable DSP for edge AI chip designers.
quadric, Inc designs and licenses a unified AI processor IP architecture aimed at edge AI chip designers. Founded in 2016, the company was established in response to the emergence of AI workloads that it argues are rendering fixed-function NPUs obsolete. Its central proposition is that neural network inference and traditional DSP and control code need not be split across multiple processors.
The company's core product is the Chimera GPNPU (General Purpose Neural Processing Unit), a processor IP architecture that combines ML accelerator performance with the full C++ programmability of a modern DSP. A single licensable processor IP handles both inference and conventional code, with deterministic timing and power efficiency optimised for single-batch latency. Performance scales from 1 to 864 TOPS. The architecture targets consumer electronics, enterprise, robotics and automotive applications, with ASIL B/D support for safety-critical use.
Alongside the hardware, quadric provides the Chimera SDK, a software stack built around a graph compiler that auto-compiles hundreds of models and imports ONNX models from PyTorch and TensorFlow. Extensibility is offered through ChiPy, a Python interface, and through C++ with an included LLVM compiler. The stated aim is a single toolchain across an entire chip line, replacing fragmented toolchains with one programmable processor IP.