In February 2019, IBM Research opened an AI Hardware Center with the goal of improving AI computing efficiency 1,000x over a decade. Over the past two years, IBM says it has met an ambitious goal of increasing computing efficiency 2.5 times a year.
IBM recently announced two key advances in AI efficiency. First, IBM will make digital AI cores compatible with the Red Hat OpenShift ecosystem. This will allow IBM hardware to be developed in parallel with Red Hat software, so that the software is ready by the time the hardware is released.
Second, IBM and design automation company Synopsys are open sourcing an AI analog hardware accelerator development kit, highlighting the possibilities analog hardware can provide. The Analog AI toolbox aims to solve the von Neumann architecture problem by performing calculations directly in memory.
According to Mukesh Khare, vice president of IBM Systems Research, the Analog AI toolkit will be available to startups, academics, students and enterprises. “They will all be able to … learn how to use some of these new features that come along with development. And I’m sure the community can come up with even better ways to use this equipment than some of us could.“, Says Hare.
The majority of this set consists of design tools provided by Synopsys. However, Arun Venkatachar, vice president of artificial intelligence and central engineering at Synopsys, said that IBM and Synopsys have worked together on hardware and software for the Analog AI toolbox.