Computing Beyond Silicon
When silicon transistors stop shrinking: which materials and principles carry the next generation of computers?
Open in the interactive tree →Silicon transistors are only a few dozen atoms wide, and further shrinking brings little extra speed. Candidates are stacked transistors (CFET), ultrathin 2D semiconductors, carbon nanotubes, optical computing, in-memory computing and reversible logic. None is in volume production, and all must compete with the manufacturing and software infrastructure built around silicon.
As of October 2026
Gate-all-around nanosheet transistors are now in volume production (TSMC N2 since Q4 2025, Intel 18A in Panther Lake since January 2026), and TSMC’s A12 process planned for 2029 will still use second-generation nanosheet transistors rather than a new material. Intel already ships backside power delivery, while TSMC’s A16 slipped to 2027. Intel’s then-CEO Pat Gelsinger said in 2023 that transistor density now doubles about every three years instead of two. Co-packaged optics is entering production for links between chips in 2026, but as an arithmetic unit optics is still research.
What is missing
- 2D semiconductors and nanotubes with good contacts, uniform quality and wafer-scale manufacturing
- Software and tools portable to new computing principles (analog, optical, neuromorphic)
- Precision and noise control in analog computing
- Economics: new fabs cost tens of billions of dollars
Becomes possible once solved
- Very energy-efficient AI
- Computers that keep getting faster after 2035
- New form factors (3D chips, flexible electronics)
Open steps
- Wafer-scale 2D semiconductors Medium AI leverageGrow uniform 2D layers such as MoS2 on production-size wafers with low defects and low-resistance contacts.
- Precision of analog computing Medium AI leverageReach digital-grade accuracy, endurance and repeatability in analog in-memory and optical multiply units despite noise and drift.
- Photonic and optical compute blocks High AI leverageDesign compact optical components and arithmetic units that survive foundry variation and scale to dense chips.
- Compilers for non-silicon hardware Medium AI leverageMake programs and AI models run on analog, optical and neuromorphic hardware without rewriting them for every chip.
Where AI could help
Medium AI leverage. AI speeds up materials, process and device search, but wafer-scale experiments and tens-of-billions fab costs set the pace.
- Optimize growth and fabrication recipes for 2D materials and contacts with Bayesian or active learning
- Screen candidate materials and devices with machine-learned simulations
- Inverse-design photonic and analog compute blocks
- Port compilers and tools to analog, optical and neuromorphic hardware
Shown so far
- In October 2021 machine-learning-guided process optimization produced wafer-scale MoS2 circuits on a 2-inch wafer, including a 4-bit full adder with 156 transistors (Nature Communications). source
Prerequisites
- Graphene & 2D Materials2004Atom-thin 2D semiconductors are a leading candidate to replace silicon channels
- 2 nm Chips & High-NA EUV2025
- Photonic & Neuromorphic Chips2026
Unlocks
- Post-Silicon Computers2040s?