Skip to content

🚀 Computer hardware frontiers — 2026

←

What chip designers and computer engineers are working on now. Each card states the open question, the evidence so far and the methods used — not results that have not happened yet.

  • Beyond 2 nm

    Open question: How much smaller can transistors get before atoms and quantum tunnelling stop them working?

    Where it stands: Gate-all-around nanosheets are in production; stacked n- and p-transistors (CFET) and atom-thin 2D materials such as MoS₂ are in research labs.

    Builds on: The MOSFET (1959), CMOS (1963) and FinFETs (2011).

    High-NA EUV lithography2D semiconductorsAtomic-layer deposition

  • Chiplets & 3D stacking

    Open question: Can many small chips stacked and joined like LEGO beat one giant chip?

    Where it stands: AMD, Intel, Apple and Nvidia ship chiplet and 3D-stacked products; the UCIe standard (2022) lets chiplets from different makers talk.

    Builds on: The integrated circuit and Moore's law.

    Hybrid bondingThrough-silicon viasAdvanced packaging

  • AI accelerators

    Open question: What is the most energy-efficient hardware for training and running AI?

    Where it stands: GPUs dominate; TPUs, NPUs in phones and wafer-scale chips compete. Data-centre power use is a growing worry.

    Builds on: The GPU (1999) and parallel computing.

    Low-precision arithmetic (8-bit, 4-bit)Huge on-chip memoryOptical interconnects

  • In-memory & neuromorphic computing

    Open question: Can we compute where the data is stored, like the brain, instead of shuttling it to the CPU?

    Where it stands: Research chips (e.g. Intel Loihi 2, IBM NorthPole) and memristor arrays show big energy savings on some tasks.

    Builds on: The von Neumann architecture — and its ‘memory bottleneck’.

    MemristorsSpiking neural networksAnalog computing

  • Photonic chips

    Open question: Can light replace electrical wires inside and between chips?

    Where it stands: Co-packaged optics links are arriving in data-centre switches; optical computing for AI is being tested.

    Builds on: Fibre optics and CMOS manufacturing.

    Silicon photonicsLasers on chip

  • Quantum hardware

    Open question: Can we build enough reliable qubits for useful quantum computers?

    Where it stands: Below-threshold error correction was demonstrated in 2024; scaling to thousands of logical qubits is the open challenge. (See the Quantum Lab!)

    Builds on: Classical logic gates — quantum gates generalise them.

    Superconducting qubitsTrapped ionsNeutral atomsError-correcting codes