As the CEO of MemryX, Inc., an innovative Edge AI startup founded by professors at the University of Michigan in Ann Arbor, Keith Kressin has been a passionate entrepreneur. Prior to joining MemryX, Keith was a SVP at Qualcomm, leading a spate of activities including the Snapdragon roadmap, technologies for application processing, company strategy, competitive analysis, and more. In an interview, he speaks on an array of topics. Excerpts:

Conception of MemryX

“Our team wanted to develop an AI accelerator that was customer friendly in terms of ease-of-use and scalability, while providing excellent technical metrics.” Adds Keith that the team wanted to develop a hardware accelerator that could be adapted to any trained AI software model, rather than needing to train and then modify the model to run well on target hardware. “So our focus from the start was on developing technology to meet a key customer need and hence MemryX was born.”

Challenges galore – AI technology

AI continues to grow in a variety of different applications and industries as it is used to make cities, companies, and hospitals smarter, help increasing manufacturing efficiencies, better support education inside and outside the classroom, and more.  “AI can add a layer of intelligence into any device that computes. Of course, every new technology is just a tool and those tools need to be used to increase to increase productivity and quality of life, rather than the limit freedoms or increase natural biases.” MemryX envisions accelerating Edge AI by providing a highly efficient and scalable solution that is simple to deploy and support.”

The key is the company’s innovative MemryX architecture.  “Our architecture was designed to natively support pre-trained AI models with high efficiency and utilization.  If one doesn’t consider model structures and framework, and only focuses on HW efficiency the result is theoretically strong, but practically weak.  MemryX architecture was designed not for white papers or theoretical achievements, but for broad customer scale and deployment.” The combination of at-memory computing, pure dataflow, and our unique memory-based fabric interconnect all support our ability to scale verified models into high volume with a very simple to use software SDK.

One trap for AI is focusing on the core technology itself, rather than the practical application of that technology throughout the entire value chain.  As in most industries, solving customer needs be in the forefront. AI is extremely complex with many facets.  It is also very exciting and dynamic field that attracts individuals with a wide variety of background and skills. The key is to stay focused on the larger picture while being extremely adaptable to market and technology changes.

Leadership traits

“It all starts with clear and transparent organization & communications.  People need to understand objectives, timelines, and be given the tools/resources to achieve well defined goals.  Teams need to be encouraged to be open and honest.”  Realistic, approachable, in-tune with all functions of the entire organization, humble, goal-centric, empowering, and one who honestly cares about the company, the team, and the individual. Keith’s leadership style is approachable with realistic expectations and stays in-tune with all functions of the entire organization. He honestly cares about the company, the team, and the individual.

“Fear in failing is usually rooted in the reaction to the failure in terms of embarrassment or punishment. However, if leaders encourage an intelligent risk-taking culture where we learn from our failures, then we grow to become a stronger organization. Further, failures from risk-taking can be minimized through proper hiring, organization, funding, and goal setting.”