Tiny Chip, Huge PotentialSep 14, 2021
At Mojo Vision, we have some of the smartest professionals in their respective fields boldly solving new and untried problems. Join us as we chat with Hardware and ASIC VP Renaldi Winoto about the small but mighty “brains” that will operate Mojo Lens.
What were you doing before Mojo Vision and what made you want to come here?
Before Mojo Vision, I was making Wi-Fi chips for cell phones and computers. During that journey, I realized how mature the semiconductor world was. The technology had become so good that it was a commodity. It didn’t matter if we're buying a processor from company A or company B—they would be comparable in terms of performance. Your ingenuity as a chip designer became less apparent; what’s the true performance gain with 1.3 GHz versus 1.0 GHz? It’s all good enough for most users.
What attracted me to the opportunity at Mojo Vision is that you must use purpose-built chips for anything to work in the lens. It's so small that the power consumption is really limited. If you can make things 10% faster, it actually matters; it's very satisfying when your contribution becomes directly apparent from a product feature point of view. And that's really attractive for me, because circuit design becomes hard again, and therefore interesting!
The “AS” in ASIC stands for “application specific.” Why does a project like Mojo Lens require an application-specific chip?
"Application” doesn’t mean a software app, but rather the purpose of a device or IC. In our case, serving information to the wearer is Mojo Lens’ application.
At Mojo, we know that application very well, and we can fine-tune what goes into the chips. We can tailor that processor design specifically for our use in terms of the power profile and performance. We have that ability to customize at the very granular, lower level so that we can manage the right performance and power trade-off specific to our application.
To answer your question, there’s no room on Mojo Lens for a big chip or a big battery. So, our chip needs to be very efficient and custom-built to do exactly what we need it to. And that, by definition, is going to be an ASIC.
Right now, all chips are flat, even ASIC chips. But your ASIC must fit into Mojo Lens’ curved, dome-shaped space. How are you accommodating this issue?
You’re right, chips are flat, and only flat. They’re thin—100-200 microns, about as thick as a piece of paper—and quite fragile. Any deflection during assembly will crack them and create issues. So to fit it into the Mojo Lens dome, we need to cut a larger chip into a few smaller chips. Now, the question we’re asking ourselves is how do we break up our ASIC to provide the best efficiency without sacrificing features?
What's the biggest challenge in designing Mojo’s ASIC chips: the physical size, the power envelope, or the computational limit?
All of the above! All three are challenging and all are of equal importance. It’s all about mapping out the landscape of the capabilities that you can do at what costs, and then threading that needle very carefully for maximum success.
Size is always a big challenge, though. Once you're at the transistor level, you're fighting with the physics; there's not a lot of elbow room. But if you address the problem at the system definition level, you have a lot more flexibility in finding the right architecture and the right balance of trade-offs.
With this constant tug-of-war going on, I’m guessing that you have to work very closely with the software team to set common goals and negotiate compromises.
You're absolutely right. Lots of people can write software, lots of people can design chips. But having the best chip or having the best software is really not going to get us there. The core strength of Mojo Vision is that we have all that hardware and software disciplines under one roof in an environment where there's open exchange of communication, and that's really what it takes to put this kind of system together. I think our success depends on being able to break this boundary and understand the system as a whole, not just the ASIC pieces.
How do you think your team’s work at Mojo will impact ASIC design over the next decade?
I view ASIC design as creating a toolbox to solve a problem. And here is a problem that people haven't looked at before. And we can bring our toolbox to solve those issues.
The semiconductor world has given us a very powerful tool: very tiny transistors. You can put it into a piece of silicon and manufacture millions of them with very low cost. And I think the manifestation of that in the last decade has been super-powerful mega-processors that can do everything.
But with a project like Mojo Lens, I think we're also showing the world that you can use transistors and silicon to enable something that wasn't possible before. Being able to design custom chips for an application enables you to define a new product category that just wasn't possible before. I think that's in a way, our biggest contribution.
Here at Mojo, we've been very fortunate. The team here has brought in a very diverse skillset in many different aspects. They're bringing a lot of excitement and they're bringing a lot of ingenuity as well to put this together. Getting as far as we have has been quite an accomplishment, and I can’t wait to see the finished product.