That's pretty awesome that you guys backed a semiconductor company. I didn't apply as I was skeptical you would invest in one... great to see YC expanding so far and fast though.
We're looking at having full simulation capabilities (with physical verification in progress) come the end of this year, with a shuttle run (prototype) tape out in Q1 next year. It's about a 3 month turn around, so we are expecting to have our first 100 prototype chips in the ~May-June timeframe next year. From there, we'll be putting together 20-25 evaluation units (consisting of 4 of our chips put together), and selling those... we have 4 order commitments already, which almost cover the cost of getting the prototypes made (though we are not reliant on that, as we now have outside funding). We'll be using TSMC (the largest pure-play semi foundry) and their 28nm process (with the very slim possibility of going down to 16/20nm if the price is right at that point).
Right now, our connections are primarily in academia and government, so they are the people we have talked to the most and have said they will put money up, and we do have some government grant funding. I do not want government and academia to be our primary focus though (Any thoughts on that?), although early non dilutive funding is nice. For the applications we have specced out that they care about, we are showing around a 10x efficiency increase for dense matrix applications (over top of the line CPUs and GPUs currently used), and even enabling some applications that were not really practical on existing systems. I can't go into too much detail on those publicly, but can talk about them offline.
When it comes to our best application space, and what I would like to target, is the signal processing and mobile basestation space. Our best performance and efficiency (25x the best CPUs, GPUs, and DSPs at the moment) comes with FFTs, and we're capable of doing next generation large FFTs better than the best DSPs are at doing the current generation types. The few people (mostly academics in the field) I have talked to in this think this is a big game changer for that industry, and the problem we have is that we don't really have any connections in the industry itself. My idea has been we should try to build up as much evidence before talking to more people in it.
While I have talked about two areas, our architecture isn't really limited to any one, it just performs exceptionally well for those... but planning for wider market adoption is I think too far ahead of us at this point.
With our current funding, we'll be able to get through making 100 (16 core) prototype chips, build those into evaluation units, and test them with customers... from there, doing a full tape out and getting the full 256 core chip to market will be another $10 to $15 million, and we expect to be able to tape out and go into mass production in the first half of 2017.
Question: Do you think the Thiel-style monopoly approach would work in our case? I initially imagined we would be able to enter and do very well in the High Performance/Supercomputing area, eat up market share from the big guys (Intel, NVIDIA, etc) who don't care about the "small" HPC market. From there, expand to large business data centers (Google, Facebook, Amazon, etc) and mobile basestation systems, and so on until we hit mass market devices. Do you think this is a decent (although very simply described for brevity) plan for the 3 to 5 year timeframe?
Are you sure the power savings for mobile base stations are worth building chips for?
I don't know the numbers, but suppose for example that there are 100,000 base stations that currently use 100 watts each doing the DSP. That's 10,000,000 watts total, costing about $10M/yr at $0.12 / kWh utility rates. If you can save 80% and capture 1/3 of the savings for the first 2 years (typical of efficiency improvement sales), you make $5.3M. If it costs $15M to design the chips, that's a bad business.
To have a good business doing energy-efficient chips, you'd need to target applications burning a billion watts from the grid, so you can make a fraction of a billion dollars.
The mobile device market values energy efficiency much higher than $0.12 / kWh. That might be a better domain to target.
I'm curious if there might be a premium brand strategy with this. If you work out the economics, a Prius almost never saves you enough in gas mileage to be worth the extra cost of the car. People still buy them anyway, because they want to advertise to everyone that they are environmentally conscious and care about the planet, and in this era of global warming, pollution, and institutionalized selfishness, that's worth something. It's conspicuous economy.
This would require a radically different marketing strategy, but what if they partnered with artists and sculptors and the advertising departments of the companies whose property the base stations would be located on? A mobile base station is usually designed to be as inconspicuous as possible; what if they flipped that around and turned a bunch of them into public art installations? The sales pitch to companies would be "Show your customers that you care about the planet, and get them talking about your brand." That's an advertising business, where they could charge way more than the power savings business.
Mobile basestations in developing markets are badass. Grid doesnt work 24 hours. Small diesel power generator is used as backup. Backup usage ranges from 3 hours a day in good electricity areas to 21-22 is worst.
Although mobile device market is certainly better target as a market.
This sounds really good. If mobile base stations are the right first customers, go try to get some of them to agree to buy your chip (if you deliver it as planned), and then go try to raise a seed round. After you deliver the evaluation units, try to raise a $10-15MM seed round.
I'd focus less on academia and government. They are usually difficult investors and difficult first customers. Non-dilutive funding is nice, but dilution is not so bad in a binary-outcome situation, and venture investors are usually far faster and more helpful.
I think taking over the "small" HPC market first is a very good plan.
I've questioned my sanity at many points due to thinking that the "small" HPC market is our best route as that is what none of the big guys are looking at, and they are horribly underserved. Most investors and others we have talked to have said to focus where the current trends/money is (mobile devices themselves, not base stations).
One more question: I'm an engineer at heart, and I want to stay involved in engineering at some level, but don't want to give up control of the direction of the company. Do you think it makes sense/is possible to be CEO and still be involved in doing day to day engineering/design work, and if so any tips on balancing that? If not, at what point does it make sense to bring in a "professional" CEO?
Thanks for the feedback! Would love to grab coffee if you're interested... feel free to send me an email.
Not only do I think it's possible to stay involved in engineering/design work, I think it's optimal, at least for the first few years. The best technical CEOs do this--they're happy to hand off most of the organization to a COO, but not product or engineering. Mark Zuckerberg did this for a long time at Facebook, and probably still does. Every time I meet with Elon Musk I'm struck by how much of the conversation is about engineering details of SpaceX or Tesla.
We're looking at having full simulation capabilities (with physical verification in progress) come the end of this year, with a shuttle run (prototype) tape out in Q1 next year. It's about a 3 month turn around, so we are expecting to have our first 100 prototype chips in the ~May-June timeframe next year. From there, we'll be putting together 20-25 evaluation units (consisting of 4 of our chips put together), and selling those... we have 4 order commitments already, which almost cover the cost of getting the prototypes made (though we are not reliant on that, as we now have outside funding). We'll be using TSMC (the largest pure-play semi foundry) and their 28nm process (with the very slim possibility of going down to 16/20nm if the price is right at that point).
Right now, our connections are primarily in academia and government, so they are the people we have talked to the most and have said they will put money up, and we do have some government grant funding. I do not want government and academia to be our primary focus though (Any thoughts on that?), although early non dilutive funding is nice. For the applications we have specced out that they care about, we are showing around a 10x efficiency increase for dense matrix applications (over top of the line CPUs and GPUs currently used), and even enabling some applications that were not really practical on existing systems. I can't go into too much detail on those publicly, but can talk about them offline.
When it comes to our best application space, and what I would like to target, is the signal processing and mobile basestation space. Our best performance and efficiency (25x the best CPUs, GPUs, and DSPs at the moment) comes with FFTs, and we're capable of doing next generation large FFTs better than the best DSPs are at doing the current generation types. The few people (mostly academics in the field) I have talked to in this think this is a big game changer for that industry, and the problem we have is that we don't really have any connections in the industry itself. My idea has been we should try to build up as much evidence before talking to more people in it.
While I have talked about two areas, our architecture isn't really limited to any one, it just performs exceptionally well for those... but planning for wider market adoption is I think too far ahead of us at this point.
With our current funding, we'll be able to get through making 100 (16 core) prototype chips, build those into evaluation units, and test them with customers... from there, doing a full tape out and getting the full 256 core chip to market will be another $10 to $15 million, and we expect to be able to tape out and go into mass production in the first half of 2017.
Question: Do you think the Thiel-style monopoly approach would work in our case? I initially imagined we would be able to enter and do very well in the High Performance/Supercomputing area, eat up market share from the big guys (Intel, NVIDIA, etc) who don't care about the "small" HPC market. From there, expand to large business data centers (Google, Facebook, Amazon, etc) and mobile basestation systems, and so on until we hit mass market devices. Do you think this is a decent (although very simply described for brevity) plan for the 3 to 5 year timeframe?