Who is Mipsology?
AI is one of the hottest areas of technology, and advances in deep learning and neural networks are opening a world of possibilities for future innovations. These deployments are poised to meet growing demand for AI: Forrester Research predicted that AI will be integral to every part of business operations by 2024.
However, there is currently a bottleneck between AI innovation demand and delivery. Industries are deploying AI, but some are finding it challenging to tap its potential for many applications. Mipsology’s technology is designed to eliminate this bottleneck and enable cost-effective, high-performance AI applications that work in every environment.
A Bit of Background
The Inference stage of Machine Learning applies the knowledge from a trained neural network model to infer a result. The system might “see, hear or touch something” and, based on its training, will create predictions about what it’s sensing. Accelerating this process is critical to many commercial and industrial AI-based applications that require quick and automated processing of vast amounts of images or video content and accurately adding value to them. However, driving computational performance during inference can be complex, highly specialized and expensive, as well as time-consuming to develop or update when changes are made.
Field-programmable gate arrays (FPGAs) are integrated circuits that excel across a spectrum of computing application use cases, including AI-based ones. They are better equipped for the inference stage than the more commonly used graphics processing units (GPUs), central processing units (CPUs) and application-specific integrated circuits (ASICs). They are hardware programmable and offer a large number of parallel computations, enabling multiple calculations and processes to take place simultaneously. FPGAs are already widely adopted in the industry as a general purpose component, and their long lifespan and ability to be reprogrammed have shown them to deliver the best hardware-level acceleration. Unfortunately, it takes special expertise to program FPGAs. While FPGA technology lends itself quite nicely to AI, FPGAs are very difficult to use and lack the level of abstraction necessary for optimal inference performance.
Introducing Mipsology
Mipsology is the first AI innovator to achieve the best inference performance with the flexibility to immediately meet the demands of a broad range of AI applications, all with “plug and play” ease of use.
In 2015, after spending two decades developing world-leading FPGA-based computers, our team of engineers saw the opportunity to develop the industry’s best compute engine for neural network inference: The ZebraTM software platform.
Zebra accelerates inference computation of neural networks on FPGAs for all machine learning applications, and completely removes the complexity of FPGAs. Zebra’s zero effort IP creates the first plug-and-play solution that unleashes the industry’s best performing AI inference accelerator, delivering long life span, broad application flexibility and lower cost of ownership.
Not all advanced neural network applications are created equal; some need speed, others low cost, and still others low latency or fast time-to-deployment. Zebra is the first and only solution that can meet these varying application demands, enabling developers to create and deploy innovative inference algorithms and new data sets in a fraction of the time (days) compared with today’s benchmarks (months, even years).
With a single command, an engineer can replace GPUs or CPUS for inference with Zebra-powered FPGAs that deliver industry-leading performance and are tuned to the critical needs of any neural network application.
Looking Towards the Future
We are pleased to report that the industry is taking notice! In 2019, Mipsology won first place in the prestigious 2019 i-Nov Innovation Contest, receiving a $2M grant from the French government. We also have several distinguished industry partners including Xilinx, Avnet Asia, Advantech, Western Digital, Technology Unlimited and Silicom Connectivity Solutions. Our platform currently supports 14 FPGA boards, and this will continue to grow because there are virtually no limitations on the FPGAs we can support.
Our deep foundation in mathematics for neural networks drives everything we do. FPGAs allow us to deliver the performance and accuracy that originate from the mathematics, but this is not the full story. Zebra has been designed to be used without extra engineering, in all locations where AI-based applications can run, and has the capacity to be the computing engine across the full industry for all AI-based applications — from smart cities to transportation and industry, and from images and video to data centers and the cloud. We look forward to working with all of you!
For additional information about Mipsology and Zebra, please contact us at zebra@mipsology.com.
