Intel has announced two non-von-Neumann processor lines—quantum and neuromorphic chips—as strategic alternatives to its lagging position in AI accelerators.
Overview
Intel’s roadmap now includes:
- Quantum processors: 12-qubit spin-qubit test chips (Tunnel Falls) and a planned 1,200-qubit system by 2027.
- Neuromorphic processors: Loihi 3, a 20-mm² chip with 240 million synapses, sampling to select partners in 2024.
Both architectures abandon traditional GPU/TPU-style matrix math in favor of physics-based computation.
What each line does
Quantum (Tunnel Falls / 1,200-qubit roadmap)
- Targets optimization, cryptography, and material-science problems.
- Runs at 1 kelvin; requires cryogenic control planes.
- Software stack: Intel Quantum SDK (C++/Python) with LLVM-based compiler.
Neuromorphic (Loihi 3)
- Event-driven, spiking neural networks for real-time sensor fusion and adaptive robotics.
- 100× lower power than von-Neumann CPUs on sparse, dynamic workloads.
- Supports Lava open-source framework; integrates with ROS 2.
Tradeoffs
- Quantum: Error rates (~1e-3 per gate) limit circuit depth to ~100 gates; no fault tolerance until 2027 roadmap.
- Neuromorphic: Limited to low-precision (1–8 bit) integer math; no support for floating-point or transformer models.
- Common: Both lack mature tooling for mainstream developers; require specialized IDEs and simulators.
When to use it
- Quantum: Early-stage R&D in pharma, logistics, or post-quantum cryptography.
- Neuromorphic: Edge devices with sub-100-mW power budgets (drones, wearables, industrial IoT).
- Not for: Training large language models or running existing CUDA workloads.
Bottom line
Intel’s pivot is a long-term hedge: quantum and neuromorphic chips won’t replace GPUs in 2024, but they offer a path to differentiate in niche markets where von-Neumann architectures hit power or latency walls.