There are at least two big missing pieces in the CHIPS Act of 2022, the US$52 billion subsidy for US-based semiconductor manufacturers that has just been sent to President Joe Biden for his signature.
The first missing piece is definitional. There is no statement in the Act of what constitutes an “advanced” semiconductor. Second, not only is there no priority on military capability, but there is also no requirement or guidance for protecting newly developed technology funded under the Act.
To begin with the question of what constitutes an “advanced” semiconductor, Intel is investing $20 billion in a new manufacturing facility, or fab, in Columbus, Ohio. The new plant intends to manufacture semiconductors with feature sizes of 5 nanometers (nm), smaller than the 14-nm used in most of today’s products; “advanced” chips are generally considered between 7-nm and 3-nm.
Intel made it clear that “no subsidy, no plant.”
Advanced chips with very small feature sizes offer very high performance and low power consumption. Computation-intensive applications, such as fast Fourier, transform iterations for submarine detection or high-speed calculations in artificial intelligence, cannot be achieved on slow general purpose processors.
China has already succeeded in making a basic chip at 7-nm, but assessments of the technology suggest it was done using older processes and the chip lacks some advanced features. It isn’t clear how quickly China will be able to sustain 7-nm chip production, though perhaps in a few years. If so, China could come online with commercial and military products even before the Intel facility is running in Ohio.
Most American companies, such as Intel and Nvidia, outsource their advanced chip production either to Taiwan Semiconductor Manufacturing Company (TSMC) or Samsung in South Korea. Both are producing 5-nm chips, and reportedly TSMC is moving to 3-nm and 2-nm classes, with production expected to start in the next few years.
And both are investing in new fab facilities in the US. Samsung is building an advanced plant near Taylor, Texas, to produce advanced logic semiconductors for mobile devices, 5G and artificial intelligence applications.
TSMC is building a new $12 billion fab in Phoenix, Arizona. Intel, along with its Ohio project, is also building two fabs in Arizona.
The Chips Act of 2022 does not address artificial intelligence (AI), nor does it identify AI’s importance to national security. Perhaps that is left to the Department of Commerce to consider as it hands out multi-billion dollar grants, though maybe not.
There is a growing need for advanced AI integrated circuits for commercial and military applications. Included here are Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs) and Graphics Processing Units (GPUs). Some of these chips are best suited for development, teaching and testing of artificial intelligence algorithms and others for actual operations.
American defense is not specified as the priority. And even for purely civil applications, the CHIPS Act puts in place no system of accountability and no requirement for what should be produced – the semiconductor companies get to decide, which may portend a shortage on the military (i.e., less lucrative) side.
But AI is critically important to the military for increasingly autonomous weapons and for rapid intelligence analytics. For example, a long-range weapon that can adjust its targeting based on its ability to sort through imagery and make sensitive decisions will increasingly be featured in new weaponry.
Related capabilities include smart countermeasures and counter-counter measures against enemy jamming. A weapon that can find its way through a thick and contested electronic jamming and spoofing environment will be needed in the future. Such weapons will also need to deflect kinetic and laser attacks while in route to their target.
Furthermore, there are no requirements or guidance for protecting newly developed technology funded under the Act. Likewise, there are no requirements for where the developers and operators of these facilities come from (other than no owners from China), nor are there any provisions for protecting the technology, including no requirements for cybersecurity, protection of intellectual property or even export of technology information.
It is likely Congress avoided the subject and was encouraged to do so by the Biden administration.
While it is true that trade in semiconductors is largely open and unrestricted, increasingly capable semiconductors that can be used in weapons ought to be controlled. The issue is how to do this.
At the start of the Ukraine war, the US imposed stiff microelectronics restrictions on Russia. There are few restrictions on China, other than Huawei (and those have been relaxed). The main products blocked to China are for advanced semiconductor manufacturing – the Dutch ASML DUV (deep ultraviolet) and EUV (extreme ultraviolet) lithography tools that etch chip designs on silicon dies. EUV is of greatest significance for 7-nm to 3-nm chips.
The export control issue is complicated by the fact that super-sophisticated AI chips can just as easily wind up in self-driving cars and autonomous vacuum cleaners as in advanced missiles or robotic tanks.
Congress failed to prioritize American defense capabilities and create a mechanism to protect the critical technologies. The great risk is that the American investment in advanced microelectronics will end up fueling China’s rise, just as US stealth technology is now part of China’s jet fighter fleet.
Follow Stephen Bryen on Twitter at @stevebryen