While there is no shortage of discussions, press announcements, or general buzz around the term artificial intelligence (AI), very few companies grasp the essence of what is required to truly realize the immense potential of enabling machines to have cognitive capabilities. IBM is certainly among those that do, and is innovating at a very fast pace in the AI hardware space.
AI is woven across just about every aspect of IBM these days, and the company’s legacy in hardware design provides deep insights into the importance of the silicon foundation to power AI applications. So, it makes sense that its renowned IBM Research organization has established an operation specifically dedicated to advancing AI semiconductors. Its AI Hardware Center is a global research hub that brings together industry, academic, and government partners with a straightforward goal: exceed the historical trend of more than doubling the efficiency of computing for AI each year and continue this until the end of the decade.
The AI Hardware Center shows that IBM understands some essential truisms about progressing AI.
First, existing CPU- and GPU-driven approaches to enabling machine learning, inferencing, neural networks, and other key ingredients to the AI recipe are insufficient and too narrow in their application. In order to expand towards fluid intelligence, AI needs a dedicated approach to end-to-end hardware development in order to proliferate. It includes new computing accelerators, technologies, and architectures designed and optimized specifically for AI computation. It also entails broad expertise in algorithms, software, system integration, and applications. This will also require an unprecedented level of performance to power the algorithms AI needs to solve the toughest problems — orders of magnitude more than we have today.
The mission of the AI Hardware Center is clear, as stated in a blog post from IBM Research:
The coming generation of AI applications will need faster response times, bigger AI workloads, and multimodal data from numerous streams. To unleash the full potential of AI, we are redesigning hardware with AI in mind: from accelerators to purpose-built hardware for AI workloads, like our new chips, and eventually quantum computing for AI.