“We are thrilled and honored to welcome Professor Mahadevan Ganapathi (“Gana”) to the SiMa.ai team. As a renowned expert in the compiler field, Gana will work closely with SiMa.ai engineering teams to advance the capabilities of our state-of-the-art ML software stack. As a first year Ph.D. student at Princeton University back in the mid-1990s, I had to make the all-important decision of choosing a thesis advisor and area of research. During this period, I studied many research papers. One particular paper that stood out to me was “Code generation using tree matching and dynamic programming” by Alfred Aho, Mahadevan Ganapathi and Steve Tjiang. I found this paper to be very fascinating, and it was one of the reasons I chose to study compilers for embedded systems for my Ph.D. thesis. As part of my thesis work, I built a retargetable compiler for embedded DSPs using some of the principles described by Gana in this paper. “
Recently I had the privilege to sit down with Professor Gana to gain his perspectives in the following areas:
- ML and Embedded edge industry
Ashok: “Your track record and breadth of knowledge is extensive as it is impressive. Please describe your background in both academia and industry, and what you will be working on at SiMa.ai”
Professor Gana: “I’ve been working on compilers for more than four decades. The first two decades I spent sharing my passion for teaching & research in Retargetable Code Generation & Code Optimization, Interprocedural Analysis and Functional Languages as a Professor of Computer Science. I have published several papers including compiler back-end technology elaborated & referenced in compiler textbooks & Advances in Compiler Technology in Annual Reviews of Computer Science. I have also directed research and served on technical advisory boards at various industries & universities, including being the Stanford Computer Forum liaison to industrial affiliates. During the past two decades, I was consulting in industry in diverse areas such as CAD, hardware verification, synthesis, silicon compilers, operating systems, computer network architectures and AI/ML compilers. At SiMa.ai I will work directly with their compiler, system architecture and hardware design teams and anywhere else my expertise is requested.”
Ashok: “Why did you join SiMa.ai, a small startup, as a consultant?”
Professor Gana: “The challenges of a small start up are really intriguing – you have an opportunity to contribute to various areas of the business, ranging from technology, business strategy, to product streams that gives you an overall better experience. I am excited and looking forward to learning and contributing to the emerging AI/ML market.”
Ashok: “What SiMa.ai value proposition do you find unique or special?”
Professor Gana: “The SoC architecture is attractive as it includes several distinct processors, and compiling software for their ISAs is a challenge that enables new opportunities and algorithms as well as AI/ML infrastructure. The AI/ML infrastructure is attractive. SiMa.ai’s ML software stack enables customers to deploy any computer vision application containing any ML network onto the MLSoCTM, which is 10x the FPS/watt of competitive products. Customers can accomplish this result within a matter of minutes, with push-button experience! My engagement is to help enable customers to easily deploy additional ML applications while further improving FPS/watt competitive advantage.“
Ashok: “What are your thoughts on ML and the Embedded edge industry?”
Professor Gana: “The embedded edge industry is a promising area as it has started helping the public. However, it requires much more work to engineer it to some standardized production quality product whereby you can iron out false positives and create a better approach for inferencing. In my opinion, this endeavor will take a couple of years. There are many players in this space who started out with promising goals but have since dropped out due to lack of funding or, simply put, their approach was not attractive. Others who are in the arena are facing the challenge of creating a solid AI/ML infrastructure and making it robust with production
quality. That Achilles heel is the system-engineering part requiring expertise.”
Ashok: “What is needed to win in this space?”
- Come up with a robust monolithic AI/ML infrastructure to address functionality, performance, precision, scalability and flexibility (retargetability).
- Create an architecture that can span several ML frameworks resulting in increased development velocity for AI models.
- As there are several processors in the SoC, a retargetable compiler framework in the AI/ML infrastructure is a dire need. It is essential to have an infrastructure that targets code for these heterogeneous architectures that are in the MLSoC.
Ashok: “What will you be working on?”
Professor Gana: “Working on AI/ML infrastructure, trying to come up with a solid infrastructure that supports a variety of AI models. I will also work on retargetable compilers that would be supported in this framework for architectures in the MLSoC. SiMa.ai already supports this function, though I will ensure that it is robust for mass production working alongside system architecture and compiler infrastructure teams.
Ashok: “Do you have any forward looking sentiments you would like to share?”
Professor Gana: “For the Embedded Edge, SiMa.ai’s software and hardware products are very helpful in all vision fronts, car industry, and anywhere cameras and sensors are available. It is an expanding market with a need for a better technology fit, which SiMa.ai provides. SiMa.ai is well positioned but there are legacy competitors in the field that we should consider. However, I believe we will win customers with our software, focusing on making it robust so that it can accept additional models, improved performance with reduced latency, lower power consumption and accommodating customers’ inferences/ watts, ideally satisfying a wide range of requirements from varying customers. This attitude will ensure a user-friendly and enjoyable experience, ultimately convincing customers to engage with SiMa.ai.”
Ashok: “Any final thoughts to share?”
Professor Gana: “Happy to be here and thrilled to be a part of the next big challenge in AI/ML.”
Ashok: “Thank you for taking the time to chat with me and welcome to SiMa.ai!”