Rethinking Semiconductor Materials: Silicon has been the workhorse of the semiconductor industry for over half a century, but as we push into the nanoscale era, new materials are gaining attention. One of the most promising is Molybdenum Disulfide (MoS₂), a two-dimensional (2D) material. Unlike silicon, which is a bulk 3D crystal, MoS₂ can form stable crystalline sheets just a single atom thick (with a molybdenum layer sandwiched between sulfur atoms). These monolayer sheets have remarkable electronic properties that make them very attractive for next-gen transistors. For AI hardware, which demands extreme performance and efficiency, MoS₂ offers several key advantages over traditional silicon.
Atomic Thinness = Better Scaling: MoS₂’s defining feature is that it’s atomically thin – roughly 0.7 nm per layer. Why is this important? In modern transistors, controlling the channel (where current flows) is critical, and thinner channels allow better electrostatic control. A monolayer MoS₂ transistor can be turned on and off more sharply, with much less leakage current, than a thicker silicon channel. This means we can potentially make transistors much smaller (in length) without the usual short-channel effects that plague silicon. Essentially, MoS₂ can scale beyond silicon’s limits, enabling continued transistor miniaturization which is crucial for packing more computation in a given area. Its ultra-thin body also means it’s naturally immune to some silicon issues like random doping fluctuations – MoS₂ doesn’t need doping in the traditional sense, simplifying transistor fabrication at tiny scales.
High Performance Potential: Despite being so thin, MoS₂ transistors have shown they can carry high currents and switch at high speeds. Engineers at CDimension have demonstrated MoS₂ devices operating in the GHz range, and with breakdown voltages exceeding 400 V for power applications. The material has an intrinsic bandgap (~2.4 eV in monolayer form) that is larger than silicon’s ~1.1 eV, which is beneficial for low off-state leakage (transistors truly turn off when they’re supposed to). In terms of electron mobility (how quickly charge carriers move), MoS₂ can achieve decent values (more than 120 cm²/Vs and improving with fabrication techniques). While silicon still slightly leads in mobility at room temperature in its bulk form, MoS₂’s mobility is more than sufficient for high-performance logic, especially given the other advantages. Additionally, MoS₂ and other 2D materials have shown the ability to withstand high electric fields – making them robust even as we aggressively scale down device dimensions.
3D Integration Friendly: One of the biggest advantages of MoS₂ for our purposes is that it’s compatible with back-end-of-line (BEOL) processing. That is, we can growMoS₂ layers directly onto wafers at relatively low temperatures (~250 °C), which will not melt or damage the underlying silicon circuits. This is huge for 3D integration. You could never grow a silicon layer on top of another silicon chip without extremely high temperatures (over 600 °C) which would destroy the chip. At CDimension, however, MoS2 can be synthesis after the silicon chip is made, acting as a “Layer 2” for transistors This means we can take a finished silicon wafer and add an MoS₂-based circuitry layer right on top of it – effectively creating a monolithic 3D IC. The ability to integrate beyond-silicon materials on silicon opens the door to complex heterogeneous chips that marry the best of both worlds (silicon’s mature platform + the new material’s capabilities).
Energy Efficiency for AI: AI workloads, especially in inference, consume vast amounts of energy. Any material that can improve energy efficiency is valuable. MoS₂ transistors can operate with 1000x lower leakage and much lower supply voltages (thanks to their excellent gate control), translating to lower power consumption. Furthermore, MoS₂ can be used to build ultra-dense memory and analog devices that complement logic. For example, an MoS₂-based memory bitcell footprint could be made far smaller than a silicon DRAM cell due to stacking, allowing more memory on-chip (and thus less energy spent accessing off-chip memory). In our high-bandwidth memory solution, MoS₂-based memory layers can achieve up to 1000× lower power per bit– a staggering improvement that directly impacts AI system efficiency. Another aspect is temperature: MoS₂’s performance doesn’t degrade as quickly at high temperature as silicon’s does (for certain device architectures), and it generates much less heat for the same operation. This means cooler chips or the ability to push performance higher within the same thermal envelope.
Industry Validation: It’s worth noting that the industry heavyweights are also looking “beyond silicon.” For instance, our Intel friends have been actively working on 2D materials like MoS₂ for future transistors, seeing them as candidates for extending Moore’s Law. Academic and industrial research in the past few years has yielded breakthroughs like high-quality wafer-scale MoS₂ growth and integration of MoS₂ transistors with silicon CMOS. All this momentum gives confidence that MoS₂ is not just a lab curiosity; it’s a practical material for the next era of electronics. CDimension’s approach capitalizes on this by being one of the first to productize MoS₂ in an AI chip context. We use MoS₂ where it makes sense – for adding layers of computing and memory – while still interfacing with conventional silicon circuits.
In summary, MoS₂ matters for AI hardware because it unlocks capabilities that silicon alone cannot provide. It allows us to build smarter chips: chips that are denser, more energy-efficient, and 3D-integrated. As AI models grow and demand more from hardware, materials like MoS₂ ensure we can keep up without an insane increase in power or cost. They represent a new toolkit for engineers. At CDimension, by integrating MoS₂ into our designs, we are essentially future-proofing AI hardware – ensuring that as silicon reaches its limit, our chips continue to deliver exponential improvements. MoS₂ and its 2D cousins could very well be to the next 50 years what silicon was to the last 50.