SiFive received a big injection of cash this week that will help the 6-year-old chip designer focus on its most advanced technology, which pits it directly against Arm’s chip designs.
Days after SiFive unloaded its custom chip unit called OpenFive to a Canadian company for $210 million in an all-cash deal, SiFive said Wednesday that it had raised $175 million in a new round of Series F financing, which values the business at $2.5 billion. The founding round was led by Coatue Management, and SiFive CEO Patrick Little told Protocol that it pushed back the funding several weeks because it wanted to ensure a few more investors were able to participate.
SiFive makes chip designs using the open-source RISC-V architecture developed at the University of California, Berkeley by the company’s founders. To Little, part of SiFive’s technological edge is its independence: The underlying technology is open source, and royalty free, and is developed by dozens of other companies — Intel included RISC-V tech in its $1 billion innovation fund. The per-watt performance of the designs — the computational horsepower achieved with one watt of energy — is roughly 30% better than the competition, according to the company.
“The performance per watt differential is extremely material and particularly in those designs where power is important,” Little said. “We found that the industry is figuring that out.”
Little said that the funding will help SiFive grow faster, and focus more on its most powerful designs for data center and machine-learning applications, which yield a greater profit and help the company grow revenue more quickly. The most advanced chips are typically the most expensive to develop, costing tens or hundreds of millions.
“At the end of every major deal, when we talk to customers, it’s us or it’s Arm,” Little said. “It really gets down to that. When the decisions are made, and it’s down to the short list, with any particular customer, in almost any vertical is really Arm and SiFive.”
Though RISC-V doesn’t command significant market share compared with Arm or x86 chipmakers such as Intel or AMD, the technology received renewed interest after Nvidia’s failed bid to acquire Arm. If Nvidia had managed to succeed, it would have given it control over core designs many of its rivals rely on, prompting many chip executives to take a close look at any alternative technologies to Arm designs.
“So many customers that are betting their roadmap on Arm’s ability to deliver, I think the [Nvidia deal] left a permanent scar,” Little said.
“Anything can happen to an [architecture] that is driven by a single company,” Little said. “Just ask the Sun guys about SPARC, the DEC guys about Alpha, ask the MIPS guys about MIPS. I just think that strategically speaking, from a risk profile, it’s dangerous to bet your entire company’s roadmap on one other company.”
But licensing chip designs that appear in billions of smartphones has proven to be a tough business for Arm. For chip companies, success is often a function of size and as semiconductor tech gets more expensive to develop, businesses are rapidly attempting to achieve scale. A larger revenue base gives a company more resources to deploy where needed, with the industry average research and development budget sitting at 18.6% for 2021, according to an industry association.
Softbank, which owns Arm, reported its fiscal 2020 adjusted profit of $596 million on revenue of $1.98 billion. The relatively small amount of revenue — Intel reported $79 billion in annual revenue, and TSMC reported $56.9 billion — presented a problem for Arm executives: In order to continue to innovate, Arm needed more resources than its current standalone intellectual property business could generate. With Nvidia backing it, Arm presumably then would have the research and development cash necessary to make a concerted attempt to take data center market share from AMD and Intel.
It’s not clear how large SiFive is at the moment, as the company declined to disclose its profit or sales. Little said that SiFive is going to circumvent that concern Arm presented by offering a licensing model — similar to Arm’s — for some of the markets it serves, but also offering additional products, including potentially chips themselves, in others. And Little likes the licensing tech as a business model because it has the flexibility of selling into just about any market — automotive, edge computing or data centers.
“Yes, the IP model is very important to us in some domains,” Little said. “In other domains we will probably be offering more solutions. It could go all the way to a chip in the future. We don’t know. But we are starting to build the value stack in hardware and software to be greater, and greater, and greater so we can capture more of the pie.”
Capturing more of that pie isn’t going to be easy, as SiFive faces entrenched customers in most of its big markets that have already poured billions of dollars into developing the best chips and software stack to power them. For Little, the answer is to make tech that is indispensable to its customers, taking advantage of any business successes to help achieve the next thing.
“We can’t be complacent, win one domain and sit on it,” he said.