COTS is government jargon for Commercial Off-The-Shelf. This means the government going out and purchasing commercial products that are available to anyone, not something commissioned specially by the government and unavailable to anyone else. It can be applied to anyone: Microsoft is a COTS supplier since the government runs Windows on lots of its machines. But it has special significance in the semiconductor market. The government used to be a major driver of semiconductor technologies back in the 1960s and 1970s, and its purchases were significant. Now, its purchases are less than 1% of the semiconductor market, meaning that most of the time they can only purchase semiconductors that are manufactured for the broad market. Integrated circuits are a mass-production technology, and with 300mm wafers and $7B fabs, it takes significant volume to move the needle. I have learned a lot about this during the year. I started in February attending GOMAC (read my post GOMAC: A Conference that Starts with the National Anthem ) which had a lot of discussion of COTS, and a lot about another big area where government and commercial activities overlap, the allocation of scarce radio spectrum. One of the challenges with advanced semiconductor is having what is called a trusted supply chain. This means a supply chain that is entirely in the US. This is getting increasingly difficult, since the most advanced foundries are mostly in Asia, and some of the most advanced packaging technologies are available only in Asia. Since volumes tend to be small, the military/government cannot usually justify doing an ASIC chip and so they are big users of FPGAs. But only Intel/Altera is manufactured in the US, and then only for their most advanced products. FPGA manufacturers never switch the foundry for an established product line as they would need to do a lot of expensive re-qualification. So old products that might have been in volume manufacturing for years are manufactured where they were when they were new. James Chew Recently, Dr Jeremy Muldavin of the US DoD came and presented at Cadence. He currently has what must be the longest job title going: Deputy Director, Defense Software and Microelectronics Assurance Activities in the Office of the Assistant Secretary of Defense for Systems Engineering. His background is 15 years with MIT Lincoln Laboratory. Cadence's James Chew introduced him, emphasizing that this is not easy for the military. They have been successful in the past but largely using techniques developed in the 1980s, based on decades of research since the 1950s to get to modern electronics. However, the commercial market has then done so much more, and so the challenge for the military is to leverage the best practices and not expensively re-invent the wheel. The military has a general cost problem, as captured by Lockheed-Martin's retired president Norman Augustine's tongue-in-cheek 16th Law: In the year 2054, the entire defense budget will purchase just one tactical aircraft. This aircraft will have to be shared by the Air Force and Navy 3½ days each per week except for leap year, when it will be made available to the Marines for the extra day. At heart, some of the cost problem is that there is just such a huge difference in price between doing a low-volume custom design (whether we are talking a chip or a jet-fighter) and buying something off-the-shelf that is already manufactured in high volume. Just think of the difference between buying a Qualcomm cellular modem versus designing a special chip for a special radio protocol, bringing it into production, and then only requiring low volumes to amortize all the design and NRE costs. Jeremy Muldavin Jeremy titled his presentation Assuring Microelectronics Innovation for National Security and Economic Competitiveness (MINSEC) . This was set up at the direction of Congress, OMB, and NSC to have the DoD develop a strategy for microelectronics. Jeremy opened with his closing summary slide. MINSEC aims to: Provide assured access and enhancement to US-based assured design, foundries, and packaging ecosystem to generate and protect intellectual property in the industrial base and retain U.S. competitiveness Invest in disruptive R&D to develop materials, devices, architectures, and design tools for complex next generation computing, strategic applications, and enhanced fabrication That's quite a heavy lift. The reason behind it is that the US military/government are feeling under threat in the technology area, for a number of reasons, the big ones being: Semiconductor and EDA technologies advance fast, and much of the way that the government does electronic design has not kept pace Some technologies are only available in Asia, or there is a long lag between their being available in Asia and in the US There are a lot of fabless semiconductor startups—but almost none of them are in the US. There are 1,000 in China, lots in Israel, some in Europe Government investment in disruptive materials, devices, architecture and manufacturing capabilities is the US severely lags the investments by other countries ($150B by China, $100B by Saudi Arabia etc) I won't reiterate the underlying trends in electronics. Have you heard that fabs are $5-15B? Or that IoT is a big thing? Or 2.5/3D packaging is in high volume manufacturing? The challenge is that the government and defence need all of the above but with higher security and even higher performance. US today: strong domestic fabless semiconductor industry (Qualcomm, Apple, NVIDIA and others dominate profits) but hardware startups falling behind foreigners IDMs at risk R&D from top universities migrating to Europe and Asia Asia today: big increase in number of fabless firms (3000 new ones in just 2017) multiple JVs and acquisitions investments of $150B in advanced microelectronics and $60B+ in STEM and universities China still 2-3 process generations behind Possible bad future outcome (from US point of view): strong global fabless semiconductor industry, US companies invest in Asia production IDMs diversify global overproduction forces fab closures best R&D is in Asia Asia has world-class microelectronics design and production capability (7 new sub-14nm foundries) trade policy to increase domestic market share (mainly China) and government fund acquisitions US loses its leadership in microelectronics One datapoint: Stanford and Berkeley wanted $350M to set up clean room, since currently they have to go to imec (in Belgium) but it was turned down. On the other hand, TSMC now offers $200M in free silicon to universities in Asia. One of the big challenges for Jeremy at the MINSEC participants is to decide what level of commercial access can be used to result in "assured" parts. In the past it was all "guns, guards, and gates" to ensure security of parts, but that doesn't work in the cyber world, nor the global supply-chain world. Short term, there is a tactical issue. Right now, 3 out of the 4 leading edge semiconductor companies (Intel, Samsung and GF) have fabrication in the US. But the reality is that the options for domestic trusted manufacture of DoD electronics are diminishing. Longer term, there is a strategic issue. Most COTS electronics used in DoD systems is manufactured overseas so is at risk from tamper. Taking a slightly wider view, the risks are the same for non-military critical infrastructure such as banking, electric power, and so on. The shift in electronic fabrication creates the potential for overseas control, and the end/slowing of Moore's Law has commercial impacts. There is no simple solution to the challenge. It is necessary to create and foster microelectronics innovation throughout the US, which involves everything from strategic alliances, incentives, access, tax policy, venture capital, EDA and more. One particular area is bring government SoP (state of practice) closer to SotA (state of the art), but under an umbrella of low-volume manufacturing: bring governement SoP to 65nm with 193i litho, and a copper BEOL multi-beam e-beam direct write litho for low-volume 200mm tools for SotA at SoP foundries plethora of miniature fabs for high-mix low-volume production market for fab recipes and low-cost mini-fabs some specialized needs (like rad-hard) will require specialized solutions At the same time, require disruptive R&D, done in a way that some of that is captured in the US, and perhaps making use of COTS parts like FPGAs. This covers materials, novel architectures (such as combining memory and logic processing on a single part), and complex design, especially software development and tools. It appears to me to be a huge plan that will require a huge investment. But Jeremy was saying that we may be able to spend 10% of what China is spending. There is clearly more to R&D than just how much money you spend. The Japanese fifth generation computer project spent a lot of money in the late 1980s with little to show for it, for example. But in semiconductor, where a fab for 5nm is probably going to have a $10B cost, and developing a process is probably going to cost at least the same again, it is not clear that commercial technology and military/government technology have grown too far apart. It is like that old adage "Fast, cheap, good. Pick any two." The government wants state of the art silicon...but only if it is manufactured in the US...and assembled with fanout packaging...and preferably with some aspects that are unique to the US based on research that hasn't been done yet. Jeremy talked about earlier programs that we funded by the government (or jointly funded with industry). VHISC (2 generations of IC technology for $918M), MIMIC (GaAs for $570M), SEMATECH ($870M + industry money). I would describe these programs as limited in their impact on US semiconductor competitiveness. VHISC produced VHDL, but the world went SystemVerilog. SEMATECH pushed 450mm manufacturing, but the world has stayed on 300mm. I don't know much about MIMIC. Jeremy did point to some other government programs in other countries: Japan through MITI and trade increased their market share and largely drove the US out of the memory business Taiwan created ITRI as a government funded R&D institute which spun off TSMC, UMC and others China, in 2015, announced plan to catch up with world standards by 2030 with $150B over ten years. Jury is still out, of course The jury is still out, too, on whatever the US decides to do, and whether the beautiful picture below ends up being more than just a picture. Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧