We believe that advances in artificial intelligence go hand in hand with advances in computing hardware. The key ideas for deep learning are over seventy years old, yet only in the last decade have we been successfully using them to solve problems with human-level performance. What changed? General purpose GPUs were developed for accelerating graphical applications, and some plucky researchers realized they could use them to train neural networks too. Given the terminal decline of Moore's Law, we think the next wave of AI innovation will be unlocked by accelerators designed specifically for machine learning.
"Improvements in compute have been a key component of AI progress... within many current domains, more compute seems to lead predictably to better performance, and is often complementary to algorithmic advances."
What we do
We design free and open source machine learning accelerators that anyone can use. Custom silicon for ML is already used by massive companies like Google, Facebook and Tesla, but everyone else has been left out. Small and mid-sized companies, students and academics, hobbyists and tinkerers all currently have no chance of getting ML accelerators customized for their needs. We aim to change that, starting with ML inference on edge FPGA platforms. Our dream is that our accelerators help those people make new applications possible that simply weren't feasible before.
Why open source?
Open source software has revolutionized the world. The device you're using to read this almost certainly makes use of dozens if not hundreds of OSS projects. Open source has been so influential because it gives end users to understand how their software works, to make changes as they see fit, and to participate in further development. This creates a virtuous cycle of feedback and improvement that benefits everyone.
Until recently, open source silicon was little more than a beautiful idea, but the rapid rise of RISC-V, backed by companies like SiFive, has provided an example of how it can be done. We aim to emulate their approach to replicate their impact. With momentum building, organizations like the FOSSi Foundation have emerged to coalesce and guide open source silicon development. Major corporate players like Intel and Google are also buying in to the idea by providing sponsorships and resources to stimulate progress. We believe that now is a great time to get involved with this movement, and we hope to do our part by bringing the benefits of open source silicon to the people working on AI.
"Aided by open source ecosystems, agilely developed chips will convincingly demonstrate advances and thereby accelerate commercial adoption."
How will Tensil make money?
Despite the benefits of open source development, many open source projects have often struggled to become self-sustaining. A typical case is that a project with thousands or even millions of users may only have a handful of core developers making the lion's share of contributions. Without a funding source, those developers may be forced to shift their time and effort to other activities in which it's easier to get paid. This problem has inspired many, many attempts to solve it, but in the last decade there have been dozens of notable successes such as MySQL, Red Hat and GitLab. Those projects have typically used a combination of dual licensing (community and enterprise editions), cloud-based managed offerings and engineering services.
Open source silicon will require different mechanisms to respond to different dynamics, but we believe there is a lot to draw on for inspiration from those commercial OSS trailblazers. You can find out more about our commercialization plans in our Roadmap. However, no matter what paid products or services we may introduce, all our open source projects will always remain open source, and we commit to supporting and maintaining them as long as there are still people who want to be involved in them.