IP licensor Blueshift Memory Ltd. (Cambridge, England) has been awarded a grant by Innovate UK, the UK government’s innovation agency.
Blueshift Memory is a fabless IP licensor, founded in 2016, that provides memory architecture IP so that large data sets and time-critical data can be more efficiently handled. The company claims this can deliver memory access speed up of a factor of 1,000 for some specific data-focused applications.
The 13-month project tha is being supported by the UK government is entitled “Research on the application of a new generation memory architecture in computer vision AI solutions for IoT devices.” Its aim is to develop a next-generation computer vision (CV) application on edge devices for the Internet of Things (IoT), based around the company’s computer architecture, which it calls the Cambridge Architecture.
Innovate UK is providing 70 percent funding of a project worth £490,000.
Computer vision typically uses artificial intelligence to analyse and interpret the content of digital images. Chips that perform this function are being developed across multiple fields including robotics, Industry 4.0, Smart Cities and autonomous vehicles. Sometimes the processing is based on high-resolution human-friendly images and sometimes on event-based machine vision systems.
In the first case the native computational work load can be high, require lots of memory and therefore consume a lot of power. Blueshift Memory is working to reduce this and its Cambridge Architecture will be demonstrated in the final FPGA design that is being developed under the Smart Grant project.
The Blueshift Memory approach differs from using graphics processing units for AI, a commonly used approach for both training and inference in the data center.
Blueshift’s project involves custom configuration of an FPGA using deep learning, and thereby optimising it for performance and power efficiency. Blueshift also plans to integrate the technology into an ASIC alongside a RISC-V general purpose processor.
“By dramatically increasing memory access speed, the compact CV AI module we are developing will open up use cases such as onboard real-time scenario analysis in body-worn cameras,” said Peter Marosan, founder and CEO of Blueshift Memory, in a statement. “It will also help us demonstrate the potential benefits of the Cambridge Architecture IP for larger system-on-chip designs for applications like high-frequency trading and in-memory databases,” he added.
Related links and articles: