Venture studios are a rarely-discussed aspect of the startup ecosystem. These organizations partner with idea-stage entrepreneurs to quickly develop company concepts, iterate on them, and build the ones that show the most promise into full-fledged startups. One of the most well-known examples is Pioneer Square Labs, also known as PSL. They mostly spin out software and services companies;such as Boundless, a software platform helping both individuals and businesses with the immigration process. As a venture studio, PSL doesn’t just provide their startups with capital and advice -- they also help with engineering, design, recruiting, finance, and operations.
More recently, a new variant on the venture studio model has been introduced by Curie Bio. They’re a venture studio focused exclusively on earlier stage biotech startups. Because of their tight focus, they’re able to offer a suite of specialized services to their startups. They help startups manage a drug development pipeline, make external lab partnerships, run preliminary trials, and ultimately get to a viable drug far faster and cheaper than the startups would have been able to on their own.
This is so valuable because biotech is a specialized, capital intensive field where many different startups need access to similar resources. And when you think about it, the same is true of semiconductors! Designing chips involves much more than coming up with an interesting and novel architecture. Every startup has to set up complex EDA tools, build fab relationships, and go through signoff and verification. But that process is pretty similar for every startup! In theory, multiple semiconductor startups could share these resources and partnerships, and get to their first working chips faster and cheaper. So, why doesn’t this exist?
Unfortunately, unlike most of my articles, I don’t have an answer for you. But I can explain why this model would work well for silicon, its potential weaknesses, and why current alternatives just don’t cut it.
Most chip development looks similar
When startups design a chip, the initial stages of the design process look fairly similar. They have to start by sourcing EDA tools from Cadence, Synopsys, or Siemens, and also get access to the design files, called process design kits, from the semiconductor manufacturers that are used to design and simulate circuits. These tools and files are incredibly complicated and difficult to set up -- most large companies have engineers dedicated to supporting these tools. But every startup needs to do it themselves, from scratch, even though the tooling is essentially the same every time.
The design process for each chip is often fairly different, as each chip has a different architecture, different function, and different needs. But the process of verifying the chip and making sure that it does what it’s supposed to do is fairly standardized. 90% of verification in the semiconductor industry is done using a framework called the Universal Verification Methodology, or UVM. UVM is designed to be highly standardized, but flexible enough to support all sorts of digital blocks. It’s attractive to large companies because the standardization makes it easy to send all of their verification work to one large verification team. Startups, however, often lack dedicated verification teams and can’t benefit from a structure like that.
Once the logical function of a chip is designed, a new set of engineers take that logic and convert it to a physical transistor layout in a process called physical design. While physical design processes, called “flows”, are often highly customized for each chip, they usually all start from a standard “baseline” flow. Startups, however, don’t have a baseline flow, and have to build their entire flow from scratch for their first chip.
Last but not least, the chip is manufactured, packaged, and tested. There are only a small number of state-of-the-art semiconductor fabs: TSMC, Intel, Samsung, GlobalFoundries, and a few others. While large companies often have relationships with these fabs that make it easy for them to get manufacturing slots, startups often struggle to make those relationships. The same is true for packaging companies, chip testing companies, and systems integrators.
Ultimately, a huge amount of the silicon design flow looks the same from company to company. While the design of every chip is unique, every chip needs EDA tools, verification, physical design, and relationships with manufacturers. These are clear places where a venture studio could add value.
In theory, a silicon venture studio could have tools and PDKs already in place for founders, with a support engineer to ensure the tools work and that founders are supported. They could have verification and physical design engineers to ensure that the chips are tested and implemented properly. Finally, a venture studio would be able to make long-lasting relationships with fabs, testing houses, and packaging services, to ensure that startups get working silicon in a timely manner. Offering all of those services to a startup, along with funding and advising, could help them get to working silicon with far, far less capital expenditure than is required today.
Is getting to first silicon the main bottleneck?
So, why doesn’t this exist yet? In my opinion, it should, but there are a number of reasons I could be wrong. I’m making one major assumption here: that helping startups get to first-silicon-success is a valuable service for a venture studio to offer. It could very well be the case that that isn’t true!
First, it could be that the bottleneck to success is coming up with a good idea. Startups could get to working silicon, and realize that nobody wants to buy it. Luckily, the venture studio model helps with that too! Software venture studios help startups come up with good ideas and validate them with market research. A silicon venture studio could do the same. With deep knowledge of the semiconductor industry, a silicon venture studio could help make sure the underlying idea for a chip startup has promise.
On the other hand, the biggest challenge for chip startups could come later in the lifetime of a chip company. Curie Bio is able to offer so much support to their investments because biotech companies have a very clear path to returning money to investors; if a drug passes clinical trials, the company that created it usually gets acquired by big pharma for a big pile of money.
The semiconductor industry isn’t like that. Companies that have raised significant capital and demonstrated revolutionary silicon, like Cerebras and SambaNova Systems, are still private many years after being founded. There’s still a chance that those companies fail -- Graphcore was acquired by SoftBank for less money than they raised. It may very well be the case that most semiconductor companies aren’t limited by their ability to get to their first working silicon, but that they’re limited by their ability to scale and grow from mid-stage startup to a public company.
Ultimately, venture studios may not be focusing on this space because they’re worried that even if a chip company is successful in the early stage, they could fail later on because they can’t scale. It’s unclear to me how significant of a risk this is. However, that hasn’t stopped some investors from focusing on early-stage chip companies. They don’t use a venture studio model, though, which makes them less useful to early stage founders in some key ways.
Existing options aren’t enough.
Those of you who are familiar with the semiconductor startup ecosystem have probably been shouting at your screen for the past few minutes. “What about Silicon Catalyst?”.
I’ve considered applying for Silicon Catalyst myself in the past, and to be honest, I don’t believe that Silicon Catalyst is particularly effective at actually accelerating the startups who go through their program. While they do have in-kind partners that ostensibly support startups, they don’t offer significant support other than simply some free tools and IP. They also don’t offer significant funding -- they only provide $100k in initial capital, which is a drop in the bucket when it comes to making chips.
At the same time, Silicon Catalyst also operates at a later stage than a venture studio would. The key application criteria for Silicon Catalyst include “differentiated technology with IP protection” and “traction with potential customers”. That positions Silicon Catalyst as an effective way for existing chip startups to get discounts on tools and IP. It’s the right service for later stage companies who have the internal talent and technical maturity to leverage those in-kind partners, but it’s not the right service for idea-stage entrepreneurs who need hands-on support.
Ultimately, Silicon Catalyst is missing the most important things that a silicon venture studio needs. With hands-on support, especially with tooling, verification, and fab relationships, as well as high-conviction capital, I think a silicon venture studio could be incredibly valuable to semiconductor founders, as well as incredibly successful. Somebody just needs to take the initiative to build it.