Following on last week's "Chips: China's Huawei May Have Found A Way Around ASML's Technology".
From Brian Potter at Construction Physics, November 20:
I am pleased to cross-post this piece with Factory Settings, the new Substack from IFP. Factory Settings will feature essays from the inaugural CHIPS team about why CHIPS succeeded, where it stumbled, and its lessons for state capacity and industrial policy. You can subscribe here.
Moore’s Law, the observation that the number of transistors on an integrated circuit tends to double every two years, has progressed in large part thanks to advances in lithography: techniques for creating microscopic patterns on silicon wafers. The steadily shrinking size of transistors — from around 10,000 nanometers in the early 1970s to around 20-60 nanometers today — has been made possible by developing lithography methods capable of patterning smaller and smaller features.1 The most recent advance in lithography is the adoption of Extreme Ultraviolet (EUV) lithography, which uses light at a wavelength of 13.5 nanometers to create patterns on chips.
EUV lithography machines are famously made by just a single firm, ASML in the Netherlands, and determining who has access to the machines has become a major geopolitical concern. However, though they’re built by ASML, much of the research that made the machines possible was done in the US. Some of the most storied names in US research and development — DARPA, Bell Labs, IBM Research, Intel, the US National Laboratories — spent decades of research and hundreds of millions of dollars to make EUV possible.
So why, after all that effort by the US, did EUV end up being commercialized by a single firm in the Netherlands?
How semiconductor lithography works
Briefly, semiconductor lithography works by selectively projecting light onto a silicon wafer using a mask. When light shines through the mask (or reflects off the mask in EUV), the patterns on that mask are projected onto the silicon wafer, which is covered with a chemical called photoresist. When the light strikes the photoresist, it either hardens or softens the photoresist (depending on the type). The wafer is then washed, removing any softened photoresist and leaving behind hardened photoresist in the pattern that needs to be applied. The wafer will then be exposed to a corrosive chemical, typically plasma, removing material from the wafer in the places where the photoresist has been washed away. The remaining hardened photoresist is then removed, leaving only an etched pattern in the silicon wafer. The silicon wafer will then be coated with another layer of material, and the process will repeat with the next mask. This process will be repeated dozens of times as the structure of the integrated circuit is built up, layer by layer.
Early semiconductor lithography was done using mercury lamps that emitted light of 436 nanometers wavelength, at the low end of the visible range. But as early as the 1960s, it was recognized that as semiconductor devices continued to shrink, the wavelength of light would eventually become a binding constraint due to a phenomena known as diffraction. Diffraction is when light spreads out after passing through a hole, such as the openings in a semiconductor mask. Because of diffraction, the edges of an image projected through a semiconductor mask will be blurry and indistinct; as semiconductor features get smaller and smaller, this blurriness eventually makes it impossible to distinguish them at all.
The search for better lithography
The longer the wavelength of light, the greater the amount of diffraction. To avoid eventually running into diffraction limiting semiconductor feature sizes, in the 1960s researchers began to investigate alternative lithography techniques.
One method considered was to use a beam of electrons, rather than light, to pattern semiconductor features. This is known as electron-beam lithography (or e-beam lithography). Just as an electron microscope uses a beam of electrons to resolve features much smaller than a microscope which uses visible light, electron-beam lithography can pattern features much smaller than light-based lithography (“optical lithography”) can. The first successful electron lithography experiment was performed in 1960, and IBM extensively developed the technology from the 1960s through the 1990s. IBM introduced its first e-beam lithography tool, the EL-1, in 1975, and by the 1980s had 30 e-beam systems installed.
E-beam lithography has the advantage of not requiring a mask to create patterns on a wafer. However, the drawback was that it’s very slow, at least “three orders of magnitude slower than optical lithography”: a single 300mm wafer takes “many tens of hours” to expose using e-beam lithography. Because of this, while e-beam lithography is used today for things like prototyping (where not having to make a mask first makes iterative testing much easier) and for making masks, it never displaced optical lithography for large-volume wafer production.
Another lithography method considered by semiconductor researchers was the use of X-rays. X-rays have a wavelength range of just 10 to 0.01 nanometers, allowing for extremely small feature sizes. As with e-beam lithography, IBM extensively developed X-ray lithography (XRL) from the 1960s through the 1990s, though they were far from the only ones. Bell Labs, Hughes Aircraft, Hewlett Packard, and Westinghouse all worked on XRL, and work on it was funded by DARPA and the US Naval Research Lab.
For many years X-ray lithography was considered the clear successor technology to optical lithography. In the late 1980s there was concern that the US was falling behind Europe and Japan in developing X-ray lithography, and by the 1990s IBM alone is estimated to have invested more than a billion dollars in the technology. But like with e-beam lithography, XRL never displaced optical lithography for large-volume production, and it’s only been used for relatively niche applications. One challenge was creating a source of X-rays. This largely had to be done using particle accelerators called synchrotrons: large, complex pieces of equipment which were typically only built by government labs. IBM, committed to developing X-ray lithography, ended up commissioning its own synchrotron (which cost on the order of $25 million) in the late 1980s.
Part of the reason that technologies like e-beam and X-ray lithography never displaced optical lithography is that optical lithography kept improving, surpassing its predicted limits again and again. Researchers were forecasting the end of optical lithography since the 1970s, but through various techniques, such as immersion lithography (using water between the lens and the wafer), phase-shift masking (designing the mask to deliberately create interference in the light waves to increase the contrast), multiple patterning (using multiple exposures for a single layer), and advances in lens design, the performance of optical lithography kept getting pushed higher and higher, repeatedly pushing back the need to transition to a new lithography technology. The unexpectedly long life for optical lithography is captured by Sturtevant’s Law: “the end of optical lithography is 6 – 7 years away. Always has been, always will be.”....
....MUCH MORE
