Thanks to our customers, Quadric has outgrown its headquarters. Effective June 18, our new address is 270 East Lane, Burlingame, CA 94010.
Our Journey
Industry's First GPNPU
Proof of Concept
Proved the innovative Chimera architecture by producing a test chip
First GPNPU
Introduced industry's first licensable GPNPU (General Purpose Neural Processing Unit)
Product Deliveries
Began delivering production-ready Chimera processor IP to customers
Third Generation
Launched third-generation Chimera AI processors with automotive safety-grade variants
With more than 1000 hardware configurations, there is a Chimera processor suited to every high-volume inference application.
Learn more about Chimera processors
Our Company, Our Markets, Our Impact
Quadric, Inc. is a semiconductor IP licensing company. We deliver the blueprints for efficient, flexible AI processors to a wide range of customers designing chips for varied applications. Our licensees are both semiconductor companies who sell to broad user bases as well as systems companies designing chips for their own internal use. We focus on solutions for AI inference in devices and on the edge – the deployment of AI models in real world situations.
We have design wins spanning the industry's widest performance range from 1 Trillions of Operations Per Second (TOPS) to hundreds of TOPS in markets as varied as automotive ADAS systems, AI PCs, industrial sensors, robotics, office automation equipment and more.
The fully programmable nature of our Chimera processor is what distinguishes Quadric General Purpose Neural Processing Unit (GPNPU(s)) from competing NPU accelerator solutions. Unlike NPUs that are programmed in idiosyncratic hardware command streams or hardware-centric assembly code, Chimera processors are programmable using common AI graph formats, C++, and Python languages that are known and used by millions of programmers worldwide, hence the tagline "democratizing AI chips".
Quadric addresses a broader Served Available Market (SAM) than competing IP licensors giving us a greater likelihood of becoming a widely adopted market leader.
Future-Proofing AI Investment
Design Once, Update Forever — Smarter ROI in AI
The two key determinants of the return on investment in an AI-enabled chip design are:
- How many AI models can run at high performance on the device, including models not yet invented
- Can your team and downstream users easily port new models, or will you be forever dependent on the NPU vendor team to respond to market changes?
Quadric Chimera processors ensure that design projects undertaken today will have a longer lifespan and serve more use cases than designs which choose fixed-function AI accelerator blocks that have to be constantly updated to handle newer models, and which provides no benefit for already shipping products.
No one can predict which new AI model will take the world by storm in 3 years, but with a Chimera processor powering your project's AI capabilities you know your chip will be ready for every change that arrives.
Reimagining AI Compute
One Architecture for Every Layer — Solving the Graph-Partitioning Problem Once and for All
Quadric is the only licensor of AI acceleration NPU solutions that:
- No Legacy Baggage
- Did not start with an existing legacy CPU, DSP or GPU and add a bolt-on matrix accelerator
- Unified Pipeline
- Provides a single processor pipeline that runs all layers of AI models – scalar, vector and matrix
The result is the Chimera AI processor core that does not suffer the graph partitioning problem that handicaps all other NPU options on the market.
Run every layer of every model on a single Chimera processor.
Understanding AI Acceleration
Artificial Intelligence (AI) enhances the functionality of devices used in many applications – autonomous vehicles, industrial robots, remote controls, game consoles, smartphones, and more. Machine Learning (ML) is a subset of the broader category of AI. By using ML models, which are trained by sifting through enormous amounts of historical data to discover patterns, devices can perform amazing tasks without being explicitly programmed.
Training Phase
ML models are created using known labeled datasets to discover patterns
Inference Phase
Trained models make predictions when presented with new, unknown data in live deployment
Because an enormous amount of computing resources are required both for training and inference, a dedicated AI / ML compute resource is now vital in most new systems to handle AI workloads.
For most high-volume consumer products, chips are designed with tight cost, power, and size limitations. This is the market Quadric serves with innovative semiconductor intellectual property (IP) building blocks. Our AI-optimized Chimera processors allow companies to rapidly build leading-edge SoCs and more easily write application code for those chips.