Summary
Highlights
TSMC manufactures 90% of advanced chips, while ASML is the sole producer of EUV lithography machines. A new startup, Substrate, claims it can disrupt this duopoly with a sub-nanometer chip printing tool that costs half as much and prints in a single exposure. Unlike traditional models, Substrate aims to build entire chip factories around this new technology.
Chip manufacturing heavily relies on lithography, with ASML's EUV lithography machine being the most complex and expensive tool ever built. It prints transistor features in nanometer scales, determining chip density and power. EUV light, with a 13.5 nm wavelength, allowed for advanced chips but requires a near-vacuum environment and costs $400 million per machine. Only a few companies, like TSMC, Samsung, and Intel, can operate these machines reliably.
As transistors shrink, printing becomes harder, necessitating multi-patterning—splitting patterns into multiple passes. This workaround increases cost, complexity, and defect risks. Even ASML's High-NA EUV, costing nearly half a billion dollars, pushes EUV to its limits, making advanced fabs ($50 billion) and wafers ($100,000) incredibly expensive, restricting access to a few large companies.
Substrate proposes X-ray lithography as an alternative to EUV. While not a new concept, previous limitations in optics and compact X-ray sources made it impractical. Substrate's breakthrough involves compact particle accelerators that fit within a factory, enabling single-exposure printing for sub-nanometer features, potentially thousand times shorter than EUV wavelengths. This promises significantly lower costs and complexity.
Substrate generates X-rays using compact particle accelerators that accelerate electrons through magnetic fields, causing them to emit intense X-ray bursts. They've demonstrated printing 12nm features with single patterning, achieving resolution comparable to ASML's High-NA EUV, with feature consistency down to 0.25 nanometers, at a fraction of the cost ($50 million). However, X-rays require reinventing photoresists, masks, and optics, and carry risks of damage and defects if not perfectly controlled. Throughput and scaling to mass production remain significant challenges.
Instead of selling machines, Substrate plans to build its own factory in the US, develop manufacturing processes, and offer foundry services, directly competing with TSMC and Samsung. This aims to create a new factory and foundry model, but faces challenges against TSMC's decades of process mastery, massive scale, and yield learning from running 30 factories and producing 1.6 million wafers monthly.
If Substrate succeeds, the dramatic cost reduction in advanced chipmaking could unlock new markets and accelerate innovation, similar to SpaceX's impact on space travel. Lower costs would allow more attempts and faster iterations, boosting compute availability for AI. Although other companies like xLight and Inversion are also working on particle accelerators for light sources, Substrate uniquely aims to replace the entire lithography step and rebuild the chip manufacturing process, potentially driving progress across all computing-dependent fields.