The Darker Side Of Consolidation

Another wave of consolidation is underway in the semiconductor industry, setting the stage for some high-stakes competitive battles over market turf and sowing confusion across the supply chain about continued support throughout a product’s projected lifetime.

The consolidation comes as chipmakers already are grappling with rising complexity, the loss of a roadmap for future designs as Moore’s Law becomes more difficult and expensive to sustain, and a flood of new markets rife with evolving standards and entirely different sets of rules.

Some industry insiders contend consolidation has never stopped, and this new wave is a continuation of a process that has been underway for several decades. But several things have changed in recent months:

• Investment in startups is rising significantly in the semiconductor space for the first time in years, particularly in such markets as machine learning, AI, automotive electronics and cloud infrastructure, creating a flood of new startups that ultimately either will be bought or go out of business.
• Interest rates are beginning to rise, which makes the cost of borrowing capital for acquisitions more expensive and adds a level of urgency to completing these deals.
• Acquisitions of larger companies continue alongside startups, and sometimes across market segments, often leading to additional sales, spin-offs or even the shuttering of some operations within those companies. (See Fig. 1 below)

Fig. 1: M&A activity in dollars. Source: CapIQ/Mentor

All of this creates a level of chaos in an industry that has become extremely efficient over the past few decades. Acquisitions can have a big impact on product support and servicing of existing technology, which is particularly troublesome for markets where devices are expected to function for 10 to 20 years.

Moreover, these issues are becoming much more common because companies are spinning off entire groups in part to specialize on what they consider their core competencies, and in part to help defray the cost of those acquisitions. Sometimes that involves selling them to buyers in other countries, where there are language barriers, or creating independent entities that don’t have sufficient capital to maintain products at the same level as before.

“The biggest problem is all of the different communications issues and variants,” said Ranjit Adhikary, vice president of marketing at ClioSoft. “Companies merge, a lot of people leave, and the knowledge base and documentation fall apart. Development either stops or is more sporadic, and access control gets compromised. You can see this with some of the big combinations, where there are global variants of IP. They don’t disclose those variants, and only some people have access to them.”

This is a big issue in markets such as mobility, for example, where costly designs are expected to produce a number of derivatives. It’s also a problem in automotive and industrial IoT, where lifecycles may be a decade or more, and where companies are expected to continually update technology throughout their lifetime with security and protocol updates. But consolidation doesn’t always turn out to be nearly as lucrative as the acquiring company expects, and not all companies want to hang onto troubled business units.

“We always assumed you could acquire an IP company and that they would do the next generation of that IP,” said Wally Rhines, president and CEO of Mentor, a Siemens business. “But to keep people around for the next turn is very difficult. There is always a new startup that can get to market faster. So with USB 2.0, we assumed everyone would just migrate to USB 3.0, and that there would be dozens of customers. But we found out there were lots of other companies developing USB 3.0, and we were expected to provide enhancements and support. Support is a killer, especially for small companies, because whenever a customer encounters a problem they blame the IP—regardless of whether the IP is the problem. The result is we were basically an extension of their design team, which was very expensive.”

Consider, for example, a company that is spun off to another company where English is either a second language, or not spoken at all. “Support becomes a key issue,” said Adhikary. “You need to manage all of the data, but if it started out in English and changes, that becomes a problem. And when you search for IP, you don’t always know what’s still being supported and what isn’t.”

One of the key issues here is rising complexity in designs, particularly at the most advanced nodes.

“With every new node, the complexity and cost go up, so you can’t afford to mess up,” said Jack Harding, president and CEO of eSilicon. “That means you can’t always buy mission-critical IP from the usual suspects if they’re not making their own chips because everyone wants to know that the IP works in silicon already. So what we’re seeing is an increasing blend of internal and external IP. For example, you may need metal buildup to make it more compatible with the rest of the design, which means you need to clean it up on the front end.”

What’s changed
Ever since the end of the dot-com era, acquisition has been the most popular exit strategy for semiconductor and EDA startups. Over the past five years, that has shifted to include established, companies such as Mentor Graphics, Arm, Freescale, Broadcom, Cavium and MIPS, as well as an increasing number of larger-value private companies. Amazon, Google, Apple, Intel and tier-one automotive suppliers have been on their own acquisition binge, as well, to the point where it looked as if there might be only a handful of potential customers for semiconductors and the tools required to develop them.

What’s different here are the dynamics of why companies are making acquisitions, which also affects what they keep and what they sell or spin off.

“In companies that use acquisitions to specialize, they can increase profitability,” said Mentor’s Rhines. “If you look at Texas Instruments, their profitability has risen since they began focusing on analog. In 2017, they showed a 43% operating profit. The same is true for NXP. In 2014, 30% of their revenue was general products. In the last five years, they’ve been selling off their standard products companies. They acquired Freescale, and their automotive revenue increased from 20% of their total revenue to 40% of revenue. So today, their operating profit has grown from the 20% range to about 30%.”

He’s not alone in seeing this shift. “Ten plus years ago, when this started to be visibly pretty active, many of us were worried about whether this was going to be the end of semiconductors, whether that was maturation, and whether fewer customers was bad,” said Aart de Geus, chairman and co-CEO of Synopsys. “All of those have actually panned out slightly differently than the worries. For starters, it was absolutely not the end of semiconductors. There was a maturation, but it’s a very good example of maturation leading to ‘techonomic’ changes, where the consolidation of some companies gave them two opportunities. One is to consolidate to become financially and economically more viable, meaning more critical mass therefore with the potential to invest more while staying profitable. Second, there is either horizontal or vertical alignment to have technical critical mass, meaning enough to invest. And what you’ve seen is a number of companies have aimed at more vertical alignment actually, because intuitively they saw a new age coming where some of the verticals need specific treatment. A prime example still in the works is the combination of Qualcomm and NXP. Conceptually, the notion of having some companies aimed at end markets makes a lot of sense. They are absolutely vertically-specific.”

This isn’t necessarily bad news for everyone. EDA has profited handsomely from a slew of new companies buying tools for new market applications. “We have helped the company make a ‘techonomic’ step forward, meaning use it to simultaneously up our game from a technology execution point of view by aligning things better, and by having more definition to our design flow so it can be re-used more, including the IP. And at the same time, economically we have ended up with a larger share. With consolidation in EDA, which actually has to do the same—be ‘techonomically’ viable both from a technology point of view and from an economic size point of view because support worldwide is very, very expensive—this is a natural trend that has kept us all on the leading edge. As you know, the leading edge keeps moving.”

Market shifts
The leading edge is also spreading out as the amount of data increases, changing the dynamics of where that data gets processed. The idea of dumb sensors at the edge of networks connected to some very powerful clouds has undergone a sharp revision over the past six months, with an emphasis on pre-processing at low power. In short, these are becoming complex, unique designs for end markets, and design activity is exploding everywhere.

“We are moving into a data-driven economy,” said Lip-Bu Tan, president and CEO of Cadence. “It’s going to be everything about data. A lot of sensors will collect data. Depending upon the application, we are moving to the intelligent edge. You will need to collect data and process it with very low power. The other side is we will scale to the cloud. The hyperscale guys are spending $80 billion this year in CapEx to build up the infrastructure. The result will be similar to what TSMC did for fabless manufacturing. Amazon, Microsoft, Google and Alibaba are building massive infrastructure to process data in a very cost-effective way. Over time, you’re going to see massive scaling of the cloud, and that will change a lot of infrastructure requirements.”

It’s also driving a lot of startup activity, which in turn will drive a lot of acquisitions. “If you go back three or four years, most industry CEOs were asking, ‘What are we going to do next because mobile is slowing down?'” said Simon Segars, CEO of Arm. “Now we have all these choices, so which one do we go and back? If you look at the machine learning space, there are all these companies getting funding to go and build machine learning chips. A few years ago, no one was putting money into semiconductor startups.”

Fig. 2: Worldwide fabless company funding, including 45 rounds in 2017 with 8 at more than $50 million and an average of $24 million. Source: Mentor, based on statistics provided by GSA, Dow Jones Venture Source.

This is particularly true in China, where the number of chip design enterprises is exploding, financed in large part by the central government’s focus on reducing its foreign trade deficit, which also plays out in financing from state-financed investment houses and local governments.

Fig. 3: IC Design explosion in China. Source: Mentor, based on statistics provided by PWC and TrendForce.

Conclusion
The amount of venture capital being invested in the semiconductor industry is rising rapidly. Not all of that money is well spent, and at some point the investments will slow and the market will consolidate even faster than today. But at least for the moment there is a huge flood of new companies, and capital is still cheap enough that companies are being acquired at a rapid pace.

The challenge is to minimize the collateral damage in what is already becoming a feeding frenzy, and the biggest point of failure is continued support, documentation and communication as those companies are acquired. The problem of “over-fishing” across the startup world seems to have passed, but with this kind of rapid turnover of companies and expertise the impact will likely be felt for years to come. And in an industry that increasingly depends on longer lifecycles in designs, that impact may be significant.

Related Stories
Chip Dis-Integration
Continued integration is no longer the natural way forward for semiconductors. What needs to happen to make it easier?
OSAT Consolidation Continues
The merger of ASE Group and SPIL alters the competitive landscape, but more changes are still ahead.
Market And Tech Inflections Ahead
Synopsys Co-CEO Aart de Geus explains why systemic complexity will be the next big challenge, and what that means for EDA and the chip industry.