View profile

Matt Roberts's Newsletter - Issue #2

Matt Roberts
Matt Roberts
After the History of Microsystems newsletter, you might be thinking I’d be writing about something other then semiconductors this time. But there’s been movement on the Canadian semiconductor front and I’d like to share some thoughts.
First, GaN Systems, a company that I worked at and that my Dad co-founded (he was CTO) raised $150M USD ($189ish CAD) from Fidelity. Congrats to the team and investors (and hopefully eventually me). This is a huge win, but keep in mind, it’s been over ten years since I was involved in raising the Series A with Rockport Capital and Chrysalix Ventures. A lesson here for everyone: semiconductor companies take a long time, even for the number one developer of Gallium Nitride semiconductors in the world.
Speaking of taking a long time…
I just saw the most recent Bond film, which was great. My favourite Bond film, though, is A View to a Kill, which features the greatest Bond villain, Maximilian Zorin.

Zorin - (Christopher "more cowbell" Walken) - Ahead of his time.
Zorin - (Christopher "more cowbell" Walken) - Ahead of his time.
In the movie, Maximilian Zorin, plans on taking control of the world. His company, Zorin Industries, has plans on cornering the semiconductor market; all the chips manufactured are in his (or in his cronies’) plants throughout the world - the only place he doesn’t control are the semiconductor plants in Silicon Valley. But he has an evil plan
To solve that problem, Zorin plans to buy oil wells all over the west coast of the U.S. and plant bombs in them, causing an earthquake and thus flooding Silicon Valley - cornering the Semiconductor market.
Cue evil laugh.
But James Bond foils the plan and throws Zorin out of his Zeppelin (yes, Zorin has a Zeppelin - this is before billionaires started space companies).
Totally outlandish, right? Not the Zeppelin part, but the cornering semiconductor market.
No one could do that, right?
35 years later, a Taiwanese company has almost done what Zorin set out to do. They control half of all the semiconductor business in the world today.
But they did it legally and without having to flood Silicon Valley. How?
If you are Apple, Broadcom, Qualcomm, you don’t actually make the chips that you sell. It’s not worth the cost of building a semiconductor fab, mastering the dark arts of making sure the chips work, or building better processing technology to keep your chips cutting edge. You outsource that manufacturing to volume players. There are a few companies that do this manufacturing: International Corporation, Silterra, United Microelectronics Corporation, Global Foundries, Samsung, and most importantly, TSMC.
Before we get there, Intel (the semiconductor company most think of) is a rare bird; today, they are one of the only`firms that both manufacture and design their own chips. And they’re losing to AMD, who makes their chips with TSMC and Global Foundries.
Remember GaN Systems that I mentioned earlier? Those chips are made predominantly in a fab that TSMC owns. GaN Systems has designed the chips for different applications, but TSMC (for the most part) is their primary manufacturing partner. And GaN Systems’ biggest competitor, Navitas Semiconductors? Well, they make their chips there, too.
TSMC out-competes everyone, by offering better products, prices, support, and services. Because they are so good at it, the entire world is their customers, and they work well with everyone. Heck, two major competitors have their chips manufactured at TSMC Fabs. You’d think they would care, but they don’t.
Well, that was until 2021…
2021 will go down in history as the year of the “Great Semiconductor Shortage.”
Now, people are wondering if this was such a good idea to mostly work with foundries located on the other side of the world.
You can read all the articles on why this took place, but here are three easy to understand causes.
  1. China has been building a three year inventory of chips after Trump made it difficult and near impossible for them to get those chips. They’ve been keeping a solid inventory in case it ever happens again (and building domestic production).
  2. Manufacturers (and/or their suppliers) canceled their purchases of chips when they believed COVID would lead to a recession. By the time they realized it would not cause a recession, their supplies and just-in-time inventory systems ran out of time. Their backlog of orders has also added to problems, as suppliers auction chips to the highest bidder, creating inflation. You can see this in finished goods (TV’s, microwaves etc.) and in automobiles.
  3. Climate Change… the monsoons in Asia didn’t arrive this year. A win for people who usually find their homes flooded, but semiconductors is a water-intensive process and Taiwan and South Korea, where the majority of manufacturing takes place, are in a drought. They’re trucking in water, because the water pipes are empty. Couple this with fires at other foundries destroying clean rooms, alongside with work-from-home orders, and you get manufacturing volume slowing.
It’s been a rough 18 months in the semiconductor business.
I’ll discuss the council and the report shortly, but the question to start with is why did they release the report last week?
Because it was throne speech time.
After a surprisingly long gap following the recent election, the speech set out the government’s priorities. But in the tech community, everyone was looking to the ministerial letters, which have yet to be released, but many expect to endorse a national semiconductor strategy.
What’s going to be in it for Canadian tech folk? Well, let’s look at what everyone else in the world is doing. Our policy is not set in a vacuum.
In America, Biden has passed an Infrastructure Bill. Except for cleantech initiatives or social programs, semiconductors are the big winner in his plan. With a headline announcement of $50 billion, the bill passed a few weeks back.
That said, the God of Semiconductors (TSMC Founder and Chair Mark Liu) says it is not enough. Biden agrees it seems and, like Max Zorin, he has a plan.
The major aspect that the media hasn’t really pointed out is that Biden is not thinking he lives in a vacuum either. His policy document “BUILDING RESILIENT SUPPLY CHAINS, REVITALIZING AMERICAN MANUFACTURING, AND FOSTERING BROAD-BASED GROWTH” (excuse the horrible title), outlines his intentions.
It states that America doesn’t think it can win alone, and wants to work with ‘techno-democracies’ to build a decentralized semiconductor fabrication supply system, with less concentration risk in any one geography. This is the great pivot away from the traditional semi supply chain, i.e., less Taiwan and less China risk.
What we need to ask: where does Canada sit in this document? We’re mentioned throughout, but not as a major player.  
But we want to be, it seems.
Three weeks back, Minister Champagne (as a reminder, he’s the federal innovation, science, and industry minister) was in Washington.
The joint circular of the meeting stated they would:
Essentially, Canada wants to be part of any US initiatives on this front. To be left out is to be left behind.
The Liberals have a chequered history regarding supporting the Canadian semiconductor space.
Remember superclusters? When Canada spent $950 million and was going to double down on our winning tech ecosystems? A chance to own the podiumWell, there was an attempt at creating a Microelectronic (Semiconductor) Supercluster that did not get into the finals. Essentially, the government thought they weren’t important to Canada’s innovative future. Oceans and protein superclusters are… but not semiconductors. 
To emphasize the point, starting in 2017, the government decided to change the NSERC funding formula (as they also put more money directly into superclusters). This had the added benefit of forcing NSERC to focus on only funding individuals.
But I’m guessing you have not heard of CMC Microsystems.
But in Canadian semi circles, it is the central repository of our semi and photonics brain trusts. They’re a nonprofit, industry and university-led organization, and would be the first stop for any government push on the semiconductor front.
They’ve also partnered with startups that you’ve heard of like Xanadu, Spark Microsystems, among numerous others. CMC claims that about 150 companies have credited their beginning to CMC over the past 30 years0. I think that number is underselling it.
While at GaN Systems, I was so impressed with the relationship we had with CMC Microsystems. Guess who we made the Chairperson of GaN Systems’ board? Former CMC CEO, Ian McWalter.
After an outcry, phone calls, and much head scratching, the government relented in early 2019 and found new funding for CMC. But just remember that while superclusters were all the rage, a group that had been doing the work in the semiconductor space for over 30 years was almost ignored to death. 
And that is a concern; the government leviathan often looks for ‘activity’ rather than actual, you know, impact. We need to keep that in mind.
By late 2019, the government had decided to step up their involvement in semiconductors. Particularly as the Trump administration began to turn the screws on Huawei.
One step was to put $5 million into the VentureLab HardwareCatalyst Initiative, which states it is the only Canadian Semiconductor Incubator. And CMC started to get new funding initiatives.
But who could the government turn to for advice on Semiconductors? Well, in May - you might have blinked and missed the press release, and accompanying coverage, that Industry Minister Mary Ng (as a reminder, Minister of International Trade, Export Promotion, Small Business, and Economic Development) launched a Canada Semiconductor Council.
“With a mandate to build and lead Canada’s national semiconductor strategy and action plan, the coalition will work towards advancing Canadian competitiveness, strengthening trade partnerships, bolstering supply chain resilience, and propelling Canada to the forefront of the US$7 trillion global semiconductor industry,” the release reads.
That’s a huge mandate to advise on the semiconductor industry to the federal government, perhaps a good sign that they were taking the space seriously. 
What was particularly surprising was one missing piece of representation in the group’s leadership: anyone who was or is a semiconductor engineer, developer, or a founder of a semiconductor company. Not hard to find, but very surprising it is not there.
The question might be, why would the Minister set up a council that has no one on it who has built or designed a chip?
Prior to the announcement of the Council, Melissa Chee (head of both VentureLab and the Council) wrote an op-ed for the Globe and Mail:
“The barriers to entry for chip manufacturing are steep, but the return on investment speaks for itself. Growing a critical mass of anchor companies that are built up to scale in Canada – rather than starting here and then exiting at low valuations – is the key to attracting billions of dollars in global investment.
“A thriving semiconductor sector also means countless patents being filed and intellectual property created. But growing Canada into an industry leader isn’t just about owning the patents; it’s about owning the podium.”
Remember that rejected supercluster for semiconductors that I mentioned earlier?
It was named the Advanced Microelectronics Supercluster, which was led by the team at VentureLab, which Melissa Chee runs. When I asked why the Semi/Microelectronics Supercluster didn’t even make it to the finals, the primary reason they shared was that no one in government wanted to build a fab.
Well, the government-appointed council leader does, as she made clear in her op-ed and now with a report from the council last week.
The report the Canadian Semiconductor Council released is long, but in multiple places makes clear its view that Canada should build a fab. The report mentions Microsystems International, which I wrote about last issue. It seems that my previous article would be a very clear warning on the results of governments trying to finance huge capital heavy tech projects, like a fab.
But they didn’t know that story. Its clear from reading their document that Microsystems was a success, we should aim to repeat.
The Semiconductor Councils proposal calls for more venture funding of companies, more funding for research, a proposed buying alliance for fab time, it a bit of a catch all of industry wishes. All good proposals perhaps, some of these services are actually offered by CMC - maybe this is a call to expand those?
But on the point of a building semiconductor fab (euphemistically described as ”Onshore manufacturing”), I want to make clear I don’t think the timing could be worse for this.
This idea is being floated at a time when more Semiconductor fabrication capacity is being built than at any time in the past 20 years. Let’s just go through the few I was able to find in a few minutes of Googling (saved you a click):
The council’s own report does mention some of these initiatives. But they do it with the sense that we should be building one, too. Instead, I think we should be watching in disbelief as more countries saddle themselves with the expensive initial cost and maintenance of fabs.
The semiconductor industry has been historically a ‘Boom and Bust’ business. Meaning, demand outstrips capacity, capacity takes time to be created (fabs takes 3-7 years to build), capacity comes online and outstrips demand, this lowers prices and lower volume/higher costs players collapse. Rinse and repeat.
The reason this hasn’t happened in the past 20 years or so is because of a company known as TSMC, who killed off oom and bust cycles. 
And while 2021 is the boom, the bust is when all that capacity comes online in the next 3-10 years.
What about the fabs we have? Should we refurbish one? Maybe.
To my knowledge, most are U.S. owned at this point.
There’s a bunch of small fabs in Canada. In the private sector, the most important one is the Bromont, Quebec facility, located next to the IBM Semiconductor fab (which only does work for them). The area has a fun little ski hill. 
Today it is owned by Teledyne (a U.S.-owned company), who bought it from DALSA, who bought it from Zarlink in 2001.
Here is your Canadian history fix: 
Zarlink was better known as Mitel Semiconductor in the 1970’s and 1980’s.
As you might remember from Microsystems International Ltd history, Mitel was started by two former Microsystems employees…
Why did they build a semiconductor fab after the mess they saw at Microsystems? Well, they didn’t.
While the Canadian federal government was losing $90 million on Microsystem Int’l ($670 million in 2021 dollars), in 1972, the Quebec provincial government (upset the MIL Lab was in Ottawa and not Quebec) decided to ‘invest’ In a semiconductor fab through the Societe Developpement Industriel (the Quebec government investment fund) and poured in over $10 million (or $63 million in 2021 dollars). The company, Siltek International Ltd (SIL), failed, and was sold to Mitel in 1976 for about $600K. Mitel successfully operated the fab for the next 25 years, for their own use, upgrading it with government incentives. 
Quebec's version of Microsystem International Ltd. - at least it failed faster.
Quebec's version of Microsystem International Ltd. - at least it failed faster.
The success of that fab in Bromont along with the lesser success of the Microsystems fab (whose fab eventually ended up at Nortel) should be a lesson: they both were successful after the government lost millions subsidizing their creations. They look successful with hindsight because Canadians did what they are good at: designing products to be used inside of them. But they were never able to justify the cost of the initial CAPEX outlay.
The Canadian Semiconductors Councils’ own report hints at this contradiction between building ‘on-shore manufacturing’ and focusing on semiconductor design, by selectively referencing a report from BCG. The better selection from that same report would be to point out the hypothetical cost of fully domestic “on-shoring” in semiconductors in the States:
“A hypothetical alternative with parallel, fully “self-sufficient” local supply chains in each region to meet its current levels of semiconductor consumption would have required at least about $1 trillion in incremental upfront investment, resulting in a 35% to 65% overall increase in semiconductor prices and ultimately higher costs of electronic devices for end users.”
That’s not a great idea for America, or for us.
The same report from BCG points out that semiconductor design accounts for 53% of all industry R&D spending but only accounts for 13% of the CAPEX and 50% of end product value. While semiconductor manufacturing accounts for only 13% of R&D but accounts for 64% of CAPEX, & 24% of the end product value.
Or in other words: if we build manufacturing, we get very little of the R&D or value add but huge capital expenditure costs. If we focus on just design in the semiconductor business we get more benefits for less cost.
The best quote of the report is this:
“A state-of-the-art semiconductor fab of standard capacity requires roughly $4-5 billion USD (for advanced analog fabs) to $20 billion (for advanced logic and memory fabs) of capital expenditure, including land, building, and equipment. This is significantly higher than, for example, the estimated cost of a next-generation aircraft carrier ($13 billion) or a new nuclear power plant ($4 billion to $8 billion).”
I’ll just leave that there.
But even if we did build a fab lets explore two things.
One is that even if we started ordering equipment today, that equipment would not arrive for 3 to 4 years. ASML one of the worlds larges manufacturers of equipment tfor semiconductors fabs has announced they are delayed until 2025/6 for new orders. Applied Materials, Lam Research, BESI - all claim delays in meeting the demand as well.
Second, as we have seen when the government starts a new program - other worthy programs suddenly find themselves starved of funding. When Superclusters were the rage CMC Microsystems found itself without cash. What funding would we have to turn off funding today to find the billion or two annually the government would need spend? A fab costs billions. Where is the cash coming from?
Instead, Canada should be preparing for a moment where we can use all those boat anchors of fabs that everyone else has built, to build Canadian-designed chips at a low cost to manufacture.
Are we prepared for that?
Well, CMC Microsystems recently started the “Fabrication of Integrated Components for the Internet’s Edge – FABrIC” program. Beyond not having money for a marketing budget to fix the name, this is a pretty huge initiative, and it got zero press. The goal is to support the manufacturing capacity of microelectromechanical systems (MEMS), integrated photonics and specialty semiconductors and quantum devices. It’s not small either - a $700 million program with $480 million from industry and $120 million from the government. It is looking for $100 million in other matching contributions from provincial governments.
The program should be fully funded and it requires no fab. 
Canada’s venture capital investment in semiconductors is probably at its lowest level in several decades. The last VC I know of with actual semiconductor development and entrepreneurial experience is retiring. A seed fund in this space would be a good idea. More investment is discussed in the report but not as directly as I (perhaps, self servingly) would have liked. 
Regardless, it is times like this where we should be concentrating our efforts on building and funding the next generation of Canadian semiconductor companies and making sure research institutions aren’t constantly begging for dollars.
I think I can say with some limited authority that the heyday of semiconductor companies in Canada was not built when there were semiconductor fabs exhausting government credit lines. The story of that Canadian semiconductor success was built on strong founders with funding, building great companies with the government listening to them. 
Did you enjoy this issue? Yes No
Matt Roberts
Matt Roberts @mattroberts

Subscribe for weekly thoughts on venture capital, tech and history.

In order to unsubscribe, click here.
If you were forwarded this newsletter and you like it, you can subscribe here.
Created with Revue by Twitter.
With ❤️ from Toronto