For more than half a century, advances in computing have been closely associated with the progressive miniaturization of silicon transistors. That trajectory is now entering a regime in which the dimensions of critical device structures approach fundamental material and quantum-mechanical limits. Rather than signaling the end of computing progress, this constraint is redirecting research toward atomically thin semiconductors, engineered interfaces, memristive devices, quantum structures, neuromorphic architectures, and three-dimensional integration. Recent demonstrations of processors containing thousands of transistors fabricated from two-dimensional materials suggest that post-silicon computing is moving beyond isolated laboratory devices toward integrated systems. The emerging frontier is therefore not simply about making conventional transistors smaller. It is increasingly concerned with controlling information through matter at the scale of atoms, ions, interfaces, and quantum states.
Silicon Is Approaching a Different Kind of Limit
The modern computing economy was constructed around a remarkably durable principle: reduce transistor dimensions, increase device density, and obtain more computational capability from each generation of semiconductor manufacturing.
This strategy produced processors containing billions of transistors, but continued dimensional scaling has become progressively more difficult. At extremely small geometries, short-channel effects, leakage current, contact resistance, thermal density, interconnect delays, fabrication variability, and quantum tunneling become increasingly significant.
The International Roadmap for Devices and Systems identifies atomically thin two-dimensional semiconductor channels as promising candidates for future highly scaled field-effect transistors. Their extremely small body thickness provides strong electrostatic gate control, potentially allowing transistor dimensions to shrink while limiting some short-channel effects that complicate conventional semiconductor scaling.
The distinction is important. Silicon is unlikely to disappear suddenly. Instead, the semiconductor industry may enter a heterogeneous period in which silicon remains an infrastructure platform while increasingly specialized materials and device concepts are integrated around it.
From Nanometres to Atomic Layers
Among the most intensively studied alternatives are two-dimensional materials, including transition-metal dichalcogenides such as molybdenum disulfide (MoS₂).
Unlike bulk silicon, these materials can form semiconductor channels only one or several atomic layers thick. Such dimensions provide unusually strong electrostatic control and create opportunities for stacking materials through van der Waals interactions without relying entirely on conventional crystal-lattice matching.
A significant transition occurred in 2025 when researchers reported prototype computer processors constructed using thousands of atomically thin 2D transistors integrated on individual chips. Nature described the demonstrations as an early view of a potential post-silicon-transistor future, while emphasizing that substantial manufacturing and scaling challenges remain.
This distinction separates contemporary research from earlier nanoscale experiments. A functioning individual transistor demonstrates a physical mechanism; thousands of integrated devices demonstrate the beginnings of an architecture.
Comparative Landscape of Post-Silicon Computing
| Technology | Physical Basis | Potential Computing Role | Principal Advantage | Major Research Constraint | Development Outlook |
|---|---|---|---|---|---|
| Advanced Silicon CMOS | Silicon transistors and nanoscale gate structures | General-purpose CPUs, GPUs and accelerators | Mature manufacturing ecosystem | Scaling complexity, leakage, heat and interconnects | Continued evolution through new architectures and 3D integration |
| 2D Semiconductors | Atomically thin MoS₂ and related materials | Ultra-scaled logic and heterogeneous processors | Exceptional electrostatic control | Contacts, variability and wafer-scale fabrication | Strong candidate for future transistor channels |
| Memristive Computing | Resistance states controlled by ionic/electronic processes | Compute-in-memory and AI acceleration | Reduces movement between memory and processors | Endurance, precision and device variation | Promising for specialized AI hardware |
| Neuromorphic Devices | Synaptic, ionic or nonlinear device dynamics | Event-driven AI and sensory processing | Highly parallel, potentially energy-efficient operation | Programming and large-scale integration | Growing experimental ecosystem |
| Quantum Computing | Superposition, interference and entanglement | Specialized algorithms and scientific simulation | Access to computational regimes unavailable to classical machines | Noise, error correction and scalability | Likely complementary rather than universal |
| Atomic/Ionic Devices | Controlled movement or states of individual atoms and ions | Memory, adaptive computing and unconventional logic | Extremely small functional structures | Stability, reproducibility and fabrication | Early-stage but scientifically significant |
Table: Comparative research framework synthesized from contemporary post-CMOS literature. Development assessments are analytical rather than commercial forecasts.
Computing May Become a Materials Science Problem
The conceptual transformation extends beyond transistor dimensions.
Conventional digital computers generally maintain a clear distinction between processing, memory, and data movement. The processor performs operations, memory stores information, and interconnects continuously transfer data between them.
For modern artificial intelligence workloads, this architecture can become energetically expensive because enormous quantities of data must repeatedly move between computational and memory subsystems.
Neuromorphic and compute-in-memory research attempts to weaken that separation.
A 2025 review of 2D-material neuromorphic systems highlighted how atomically thin materials could support synaptic devices, dense crossbar structures and monolithic three-dimensional integration. Separate experimental work has demonstrated MoS₂-based transistor-memristor structures designed to perform synaptic computation within compact arrays.
The implication is substantial: future computing performance may depend as much on how materials physically store and transform information as on transistor switching frequency.
When Atoms and Ions Become Computational Variables
At still smaller scales, researchers are investigating systems in which ions and atomic-scale interfaces participate directly in information processing.
Research published in 2026 examined graphene-based ionic transistors in which confined ions interact with electronic states inside extremely narrow channels. Simulations indicated that collective ionic behavior and electron–ion interactions could influence switching and transport characteristics relevant to low-energy neuromorphic hardware.
Other research has used single-atom engineering to alter the behavior of memristive interfaces. One 2026 study employed isolated nickel atoms anchored to reduced graphene oxide to modify interfacial properties and ionic migration in a perovskite memristor.
These experiments illustrate a deeper shift in computational engineering.
At conventional scales, individual atoms are treated statistically as constituents of materials. At the post-silicon frontier, the position, charge environment or movement of atoms and ions may itself become part of device functionality.
Artificial Intelligence Is Accelerating the Search
AI provides an unusually strong economic and scientific motivation for unconventional computing.
Large neural networks rely heavily on matrix operations and repeated memory access. Conventional CMOS GPUs remain exceptionally capable, but scaling AI through increasingly large digital processors can impose substantial energy and data-movement costs.
This is encouraging research into hardware in which computation occurs closer to memory.
In 2026, researchers reported MoS₂-based neuromorphic hardware using signal-folding techniques intended to balance computational precision and energy efficiency in vector–matrix multiplication. Other work demonstrated protonic nickelate networks capable of combining nonlinear temporal processing with programmable memory within a common material platform.
Such systems do not necessarily replace GPUs. Instead, they point toward a heterogeneous computing ecosystem in which conventional processors coexist with specialized physical accelerators.
Key Findings
- Silicon scaling is becoming increasingly constrained by physical and engineering effects, but this does not imply an immediate end to silicon computing.
- Atomically thin 2D semiconductors provide a potential pathway toward stronger electrostatic control at extremely small device dimensions.
- Prototype processors containing thousands of 2D-material transistors indicate progress from individual experimental devices toward integrated computing systems.
- Neuromorphic and compute-in-memory architectures could reduce the energy cost associated with repeatedly transferring data between processors and memory.
- Atomic-scale engineering increasingly allows ions, interfaces and individual atomic species to influence switching, memory and computation directly.
- The post-silicon era is likely to be heterogeneous rather than dominated by a single successor technology.
The Manufacturing Problem Remains Formidable
Scientific feasibility does not guarantee industrial adoption.
Atomically thin materials must eventually be manufactured across large wafers with extremely low defect rates. Electrical contacts must remain reliable. Device characteristics must remain consistent across billions or potentially trillions of structures. New materials must also integrate with existing semiconductor processes without introducing unacceptable cost or contamination.
The same difficulty affects neuromorphic and memristive technologies. Device-to-device variability, endurance, precision and integration complexity remain important barriers. Contemporary oxide-based neuromorphic research, for example, continues to identify scalability, reproducibility and compatibility as central challenges even as increasingly sophisticated multifunctional devices are demonstrated.
Consequently, the decisive breakthroughs may come not from discovering a single extraordinary material but from developing manufacturable material systems.
A Post-Silicon Future May Still Contain Silicon
The phrase “post-silicon” can therefore be misleading if interpreted literally.
The more plausible trajectory resembles the evolution of computing architecture itself: specialization layered upon a durable foundation. Silicon CMOS could continue to provide control logic, general-purpose computation and manufacturing infrastructure, while 2D semiconductors, advanced memories, photonic components, quantum processors and neuromorphic accelerators perform specialized functions.
Even conventional silicon may acquire unconventional capabilities. Researchers have demonstrated that standard silicon transistors, when operated under unusual electrical conditions, can reproduce synaptic and neuronal behaviors, illustrating that architectural innovation does not always require abandoning established materials.
Conclusion: Computing Becomes the Engineering of Matter
The first computing revolution was largely architectural. The semiconductor revolution made it electronic. The coming transition may increasingly make computing atomic and material-specific.
As conventional dimensional scaling becomes harder, researchers are learning to manipulate properties that earlier generations of engineers attempted to suppress: quantum effects, ionic motion, nonlinear material responses, interface states and atomic-scale interactions.
The resulting machines may look fundamentally different from the universal processors that defined previous decades. Some could compute inside memory; others may integrate sensing and processing within the same atomic layers. Still others may exploit quantum states or controlled ionic dynamics.
The central question of the post-silicon era is therefore no longer simply how many transistors can fit on a chip. It is becoming a more fundamental scientific problem: how efficiently can matter itself be engineered to represent, transform and preserve information?
If that transition succeeds, the next major increase in computing capability may not originate from another conventional transistor shrink. It may begin where traditional semiconductor engineering once ended—at the scale of individual atomic layers.
