In our post yesterday we mentioned that one of the problems Arm faces is its pricing model. We had a few questions about that and wanted to elaborate on them and use that as a tool to explore the larger problem haunting semiconductor start-ups – the high upfront cost of IP licenses.
Intellectual property (IP) licenses are typically structured in two parts – an up front license payment and recurring, per-device royalty. The upfront portion is heavy, typically several million dollars, sometimes tens of millions. The recurring royalty portion varies significantly from a few cents per device up to several dollars.
This can cause wrinkles for companies like Arm. For example, several years ago, Arm started a big push to play in the Internet of Things (IoT). The appeal for Arm was straightforward – billions of IoT devices coming on the market in a huge range of industries, a large new market for them.
This presents some obvious problems for a traditional IP licensing model. If you are producing smartphones and order chips in batches of tens of millions, a multi-million upfront payment gets amortized with little disturbance. But if you are producing a small batch of devices, or the end-device is a $10 remote sensor, those upfront costs are prohibitive. In Arm’s case, they eventually realized the problem and issued a new pricing model. From what we could tell from public materials, this new structure was not much better, largely deferring upfront payments until later dates. The outcome of this new plan can best be measured by the fact that a few weeks ago Arm announced they were spinning off their IoT business to parent Softbank, not an indicator of success.
Our sense is that the cost of Arm licenses prompted a lot of IoT companies to eschew Arm entirely for chips with much less processing power. The vast majority of IoT devices do not need heavy duty processors, but a more creative pricing model might have encouraged some form of upgrades. Instead, it looks like Arm has lost large portions of this market.
By contrast, another large IP licensor ended up with a different approach. In the mid 2000’s, it was becoming clear that there were hundreds of Chinese companies looking to enter the smartphone market. As these companies looked to expand beyond China and as 3G became widespread globally, they ran into Qualcomm’s licensing requirements. It took a few years, a couple lawsuits, and ultimately the intervention of the Chinese government, but Qualcomm ended up with a much more flexible and robust licensing program in China. That ultimately led to the growth of a handful of major Chinese handset vendors who remain important Qualcomm customers. This was not a simple or painless process, but it did work out in the end.
IP licenses are a tricky issue. The companies developing that IP have to do an immense amount of work to build their solutions. No one argues about the fact that Arm cores are immensely useful and important. But then how do they price things? The standard model is a major disincentive to new companies entering a field. But if you lower prices for start-ups, the big companies – aka your biggest customers – ask for discounts too. There are no perfect solutions, just a series of trade-offs. Arm’s decision pattern seems to have hurt its long-term prospects for IoT.
The problem is that Arm is not the only license holder seeking payment for its IP. The EDA design tools companies present another major cost for new entrants to the semiconductor market. These companies like Mentor Graphics and Synopsis, provide tools for designing chips, but also license important, useful blocks for chips as well, for things like memory and I/O interfaces. Their pricing models are varied, but again there are similar issues where new companies face millions of dollars of tool and IP costs. This market actually looks a lot like the cable TV market where you sign up for ESPN but have to pay for the Hallmark Channel too.
Again there is no easy solution. The EDA companies do a lot of work that is important to the industry and deserve to be compensated for it. On the other hand, we know many chip start-ups who have to struggle to raise money only to spend a large portion of that capital on tools and IP libraries. This understandably presents a big problem for venture capital investors who are footing the bill for that IP. So it is not a surprise that one of the most active semiconductor VCs runs an EDA company as well.
Someone needs to build a Y-Combinator for chips. A start-up accelerator that has some negotiating leverage with the major IP holders, sufficient to provide an ‘umbrella’ license covering its portfolios up to a certain size. The goal should be to get a start-up to the point where it can build a test chip, sufficient to demonstrate its capabilities to customers who could then sign up and help finance a commercial part.
This will be easier to say than to actually put in place, but it is possible. But it is time that someone gets creative in solving this particular logjam.