Link here.
However, the tool that Substrate is developing does not appear to be
compatible with existing equipment and production flows, so the company
will have to reinvent the whole supply chain to be successful. However,
Substrate does not plan to sell its tool, but to build its own fab and
provide foundry services.
As integrated circuit features are getting smaller, chipmakers are using
increasingly intricate lithography tools that now cost around $235
million for an ASML NXE:3800E Low-NA EUV scanner or around $380 million
for an ASML EXE:5200B High-NA EUV scanner. As a result, fabs are
becoming increasingly expensive to build, and chips are becoming more
expensive to produce.
Substrate models that a leading-edge fab will cost around $50 billion by
2030, leaving semiconductor production to a handful of companies with
very deep pockets. Furthermore, such fab expenditures are expected to
increase the cost of a 300-mm wafer, which it claims could balloon to
$100,000 when using leading-edge fabrication processes. This will make
the development and production of advanced chips prohibitively expensive
for small companies. Substrate intends to change that and reduce wafer
pricing to just $10,000 by 2030.
"At Substrate, we have a pathway to reduce the cost of leading-edge
silicon by an order of magnitude compared to the current cost-scaling
path we are on," a statement by the company reads. "By the end of the
decade, Substrate will produce wafers closer to $10,000, not $100,000."
Note
that Substrate is by no means the only company exploring particle
accelerators as light sources for EUV or beyond-EUV lithography. In the
U.S. alone, two companies — Inversion Semiconductor and xLight — as well as researchers at Johns Hopkins University, have revealed that they are working on lithography systems harnessing particle accelerators over the past 12 months. Chinese scientists and Japanese researchers are also testing particle accelerators for semiconductor production.
Substrate's x-ray lithography.
Substrate is developing a new type of lithography system that uses a
particle accelerator to produce short-wavelength X-ray radiation (or
light) for chipmaking. The goal is to replace ASML's expensive EUV
lithography scanners with compact, low-cost machines capable of printing
transistor patterns at a 2nm-class process technology (or even more
advanced, the company claims). The firm claims the machine should reduce
chip production costs by 10 times by the end of the decade.
At the core of Substrate’s technology is a custom particle accelerator
which propels electrons (produced by an unknown emitter) to near the
speed of light using radio-frequency cavities.
As these electrons pass through sporadic magnetic fields, they gain
kinetic energy, accelerating to speeds very close to the speed of light
(a relativistic speed), which allows them to produce special types of light when manipulated. These fast-moving electrons fly through a series of magnets that flip back and forth, wiggling the electrons and causing them to release their energy and produce coherent bursts of intense x-ray light (or radiation).
That light is 'billions of times brighter than the sun,' likely
producing pulses intense enough to achieve the desired resolution and
dose. The X-ray pulses are then focused by 'a succession of perfectly
polished optics' to project a photomask onto a photoresist-coated
silicon wafer. Note that Substrate never mentions reticle and resist in
its official description, only claiming that 'bright pulses of light'
are collimated and transported 'all the way to the silicon wafer,' which
implies maskless direct-write lithography, which is good enough for
research purposes, but orders of magnitude too slow for the mass
production of chips. However, this remains speculation on our part.
In
fact, Substrate's description of its technologies is very brief and
lacks detail (perhaps for competitive reasons), making it difficult to
analyze. However, since the company mentions X-ray, we are dealing with
electromagnetic radiation with wavelengths ranging from 0.01nm to 10nm
and energies from about 100 eV to 100 keV. The shorter the wavelength,
the finer the structure that can the printed with improved accuracy, but
the harder the light is to manage and work with.
Given Substrate's achievements so far, we are likely dealing with soft
X-rays (wavelengths of 1-10 nm, lower energy) rather than hard X-rays
(wavelengths of 0.1-1nm, higher energy).
Since short-wavelength light (including EUV and X-rays) is strongly
absorbed by most materials, managing it requires a set of perfectly
polished mirrors that reflect light at grazing angles (to avoid
absorption), ultra-precise alignment, and vacuum environments. Also,
X-ray lithography requires all-new resists that can handle high-energy
photons without damage or blur.
2nm-like CD and T2P spacing
To
prove that its XRL method works, Substrate has shown off images of a
random logic contact array with 12nm critical dimensions (CD) and 13nm
tip-to-tip (T2T) spacing printed with high pattern fidelity, as well as
random vias with a 30nm center-to-center pitch, possessing superb
pattern quality and critical dimension uniformity. If such metrics could
be achieved for mass production today, this would largely revolutionize
the lithography industry, as it would enable scaling across both axis
at 2nm-class nodes (and lower) without using multi-patterning.
Modern EUV scanners with 0.33 NA optics can achieve critical dimensions
of 13nm–16nm in high-volume manufacturing, which is sufficient to print a
26nm minimum metal pitch (good enough for 2nm or 3nm-class process
technologies) and a 25nm T2T interconnect space with a single exposure.
Such disproportions emerge because chipmakers tend to optimize
resolution in the Y direction (CD) to get the tightest metal-pitch
line-space pattern, but at the cost of resolution in the X direction,
which means that T2T prints poorly or inconsistently, leading to
bridging defects, stochastic defects, yield loss, complicated design
rules, and slower scaling. To mitigate this and avoid blurred or
inconsistent line ends at tip-to-tip spacing, Intel applies
pattern-shaping tools in the X-direction with its 18A fabrication
technology, but this complicates the overall production flow and does
not fundamentally solve the issue.
Substrate's tool (assuming these are real lab results, not a simulation)
can already outperform existing Low-NA EUV scanners in terms of
achievable CDs with single-resolution patterning, and it leaves them
behind dramatically when it comes to T2T spacing printed with high
fidelity. This means that Substrate's X-ray lithography tool could
possibly replace costly EUV multi-patterning used for sophisticated 3nm
and 2nm-class process technologies or pattern shaping used for Intel
18A.
Our friends at SemiAnalysis
have managed to get more performance claims from Substrate, which look
even more impressive. The company claims it has achieved overlay
accuracy of under 1.6nm, full wafer critical-dimension uniformity (CDU)
of 0.25 nm, line edge roughness (LER) of under 1nm, and local critical
dimension uniformity (LCDU) below 1.5 nm.
If
accurate, this performance would match or surpass ASML's Twinscan
NXE:3800E in uniformity, though its overlay precision is slightly worse
than the 0.9nm machine-matched overlay standard in the latest EUV
scanners. Also, the line-width uniformity of contacts on an image
provided by Substrate is rather poor.
Assuming
the results presented by Substrate are real and achieved in a lab
environment, this means the company has solved three critical challenges
with X-ray lithography. First, build a light source featuring an
electron gun and a particle accelerator; second, create a
grazing-incidence mirror system to reflect and focus X-rays at very
shallow angles; and make the whole thing compact enough to fit into a
lab.
However, Substrate still has a lot of work to do, turning its X-ray
lithography technology from a lab success into a viable production tool.
Substrate must prove that its X-ray lithography system can maintain
beam stability, optical precision, resist compatibility, overlay
accuracy, and commercial throughput simultaneously, something no X-ray
platform has ever achieved.
Existing photoresists are incompatible with X-ray radiation, as they are
optimized for EUV radiation with considerably lower photon energy. So,
Substrate will have to invent a proper resist and then produce it at
volume. The company will also have to develop photomasks that can
sustain X-ray radiation. Grazing-incidence mirrors for X-rays are also
not in mass production, and it is unknown whether they can be
mass-produced cheaply and reliably by existing producers like Zeiss.
Substrate will also have to ensure that X-rays do not damage the
underlying transistors or introduce stochastic defects. Achieving
overlay accuracy below 1nm (to match ASML's production-level alignment
precision) remains another challenge for the company. This is perhaps
because the company still has to address issues such as wafer handling,
stage repeatability, and other factors related to high-precision
mechanics, which ASML has taken decades to solve.
Beyond
that, the tool must reach commercial throughput and consistent yield,
something that took years for ASML's EUV tools. In fact, ASML's EUV journey timeline
is quite exemplary: it has taken the industry 12 years to evolve from
an alpha demo tool (2006) to mass production (2018), and about seven
years to go from the first pre-production system (2010) to a
mass-production-capable scanner.
Speaking of mass-production-capable X-ray lithography tools, it is
important to note that Substrate has no intention of selling them to
third parties such as Intel or TSMC. Instead, Substrate plans to build
its own fabs in the U.S. (a move that could give the company
geopolitical importance in the eyes of the U.S. government), install
additional tools, and offer foundry services, thus challenging existing
chip contract manufacturers.
However, this strategy adds complexity and cost. Constructing even a
single high-end semiconductor fabrication plant would require tens of
billions of dollars in investment and a large ecosystem of suppliers and
service infrastructure, which currently does not exist for X-ray
lithography production.
Substrate would also need to integrate its XRL litho machines with
hundreds of other tools in the fab, or persuade its suppliers (such as
Applied Materials, KLA, Lam Research, etc.) to help it do so, which
likely involves further investments from the company, making its first
fab particularly expensive.
Also,
running both a toolmaking activity and a chip foundry would stretch
Substrate's technical and financial resources, which would make it
particularly hard to achieve its promised per-wafer price of $10,000 by
the end of the decade, as its investors will likely demand returns after
pouring tens of billions of dollars into the company.
However,
if Substrate succeeds in both roles, it could shift the balance of the
semiconductor supply chain back to the U.S., as the company will likely
outpace ASML's tools in terms of resolution and performance, and TSMC in
terms of design cycle time and potentially volume.
