Video by Robert / Chip Design Engineer — Exploring What Comes After Silicon
Quick Summary: Silicon has dominated computing for 60+ years, but physics is finally catching up. The industry has survived through tricks (FinFETs, nanosheets, CFETs, Power Via), but each fix buys only a few more years. Now, for the first time, a clear successor has emerged on IMEC's official roadmap: 2D semiconductors like molybdenum disulfide — materials just 3 atoms thick. A Chinese RISC-V processor with 6,000 working transistors at 99% yield proves it's real. The endgame isn't just replacing silicon — it's stacking logic layers vertically for true monolithic 3D computing.
For decades the industry shrank transistors to make chips cheaper and faster. That strategy has hit three walls:
| Problem | What Happens |
|---|---|
| Quantum tunneling | Electrons behave like waves — they leak through barriers they shouldn't cross |
| Cost explosion | Each new node is exponentially more expensive to develop |
| Thermal limits | Denser transistors generate more heat; data-center power is already a crisis |
| Innovation | How It Works |
|---|---|
| FinFET | 3D raised channel — better gate control than flat transistors |
| Nanosheets / GAA | Horizontal bars wrapped by gate on all sides — Intel's 18A node |
| Power Via | Power delivered from the backside of the chip, freeing up top-side routing |
| CFET | Two transistors stacked vertically in the space of one — coming at A7 (~2033) |
Molybdenum disulfide (MoS₂) is a 2D semiconductor — one layer of molybdenum between two layers of sulfur. Total thickness: ~0.5 nm (3 atoms).
| Property | MoS₂ | Silicon (typical fin) |
|---|---|---|
| Thickness | ~0.5 nm | 5-10 nm and thinning |
| Channel control | Atomic precision — full gate control | Leakage increases at small scales |
| Operating voltage | ~0.3-0.4 V (target) | ~0.7-0.8 V |
| Energy per switch | Up to 1000× less (published) | Baseline |
AI is driving an energy crisis — data centers are becoming some of the largest electricity consumers on the planet. The industry doesn't just need more compute. It needs more compute per watt. 2D semiconductors deliver exactly that through dramatically lower operating voltage.
Built in China, this processor contains 6,000 transistors, each with a 3-atom-thick MoS₂ channel.
| Metric | 2D Processor | Intel 4004 (1971) |
|---|---|---|
| Transistors | 6,000 | 2,300 |
| Channel thickness | ~0.5 nm (3 atoms) | ~10 µm |
| Architecture | Open-source RISC-V | 4-bit |
| Yield | 99% | N/A |
At first glance 6,000 transistors sounds tiny (vs 300B in an NVIDIA Rubin GPU), but this is a functional, instruction-executing processor — not just a lab demo. It runs real programs.
For a decade researchers showed individual 2D transistors. Getting thousands to work together at 99% yield is the breakthrough that changes the conversation from "can it work?" to "can we manufacture it at scale?"
Silicon has a 60-year head start. Growing perfect MoS₂ across an entire wafer is extraordinarily hard:
| Challenge | Detail |
|---|---|
| Compound material | Every Mo and S atom must land in the right place at the right time |
| Grain boundaries | Tiny crystals collide during growth — defects appear like cracks in ice |
| Temperature | Traditional methods need ~1000°C — hot enough to destroy underlying transistors |
CDimension claims it can grow MoS₂ at just ~200°C — low enough that the silicon layer underneath survives. This is critical because it unlocks:
| Approach | What It Is |
|---|---|
| CFET | Two transistors stacked — "add another floor to the transistor" |
| Logic folding | Stack separately manufactured silicon layers with hybrid bonding |
| Monolithic 3D | Build transistor layers directly on top of each other inside a single chip — no bonding interface |
Monolithic 3D is the endgame: a skyscraper of logic layers with shorter connections and far less energy wasted moving data between layers.
2D semiconductors won't appear in your laptop's CPU first. They'll enter through specialized markets:
| Application | Why First |
|---|---|
| RF switches (6G) | Don't need billions of perfect transistors |
| Photonics | Benefit from 2D material's optical properties |
| MEMS devices | Easier proving ground |
Every semiconductor breakthrough went through this pattern: silicon photonics, superfins, EUV lithography — all spent years in the valley before succeeding.
| Concept | Key Point |
|---|---|
| Why now | Silicon scaling has reached physical limits; quantum tunneling breaks smaller transistors |
| The material | MoS₂ is 3 atoms thick, needs ~½ the voltage, up to 1000× less energy |
| Proof it works | 6,000-transistor RISC-V processor at 99% yield — first real 2D computer |
| Manufacturing | CDimension's 200°C process makes monolithic 3D feasible |
| The endgame | True 3D computing — stacking logic layers vertically, no silicon needed |