Advanced Computing β Beyond Silicon
Classical silicon-based computing is approaching fundamental physical limits β transistor miniaturization, energy walls, and the end of Moore's Law as we've known it. Meanwhile, radically different computing paradigms are advancing in parallel: quantum computers manipulating entangled qubits, neuromorphic chips mimicking brain architecture, photonic processors using light instead of electrons, and molecular systems harnessing DNA and chemistry to compute. This page maps the key domains and tracks the most significant, sourced milestones.
Automation Progress
Quantum computers exploit superposition and entanglement to solve certain problems exponentially faster than classical machines. The central challenge has shifted from building more qubits to achieving fault-tolerant, error-corrected logical qubits. 2024β2025 saw historic breakthroughs: Google crossed the error-correction threshold, Microsoft unveiled topological qubits, and IBM charted a path to 100,000 qubits.
Google Willow: first chip to break the error-correction threshold
Google's 105-qubit Willow processor demonstrated that adding more qubits reduces rather than increases errors β crossing the "below threshold" barrier first theorized in 1995. Willow completed a random circuit sampling task in under 5 minutes that would take the world's best supercomputer 1025 years. Published in Nature.
Microsoft Majorana 1: world's first topological qubit chip
Microsoft unveiled Majorana 1, the world's first quantum processor powered by a topological core architecture. Built on a novel "topoconductor" material, it demonstrated hardware-protected topological qubits that are smaller, faster, and digitally controllable β with a path to scaling to 1 million qubits on a single chip. Published in Nature.
IBM delivers new processors and roadmap to fault-tolerant quantum
IBM unveiled fundamental progress including new quantum processors, software, and algorithm breakthroughs. The updated roadmap targets quantum advantage by end of 2026 and a 10,000-physical-qubit "Starling" system by 2029 using LDPC error-correcting codes, with a long-term vision of 100,000 qubits by 2033.
Neuromorphic computing mimics the architecture and signaling of biological neural systems β using event-driven spiking neural networks instead of synchronous clock cycles. The promise: orders-of-magnitude better energy efficiency for AI inference, robotics, and sensory processing. Intel's Loihi line and academic efforts like Tsinghua's Tianjic are pushing the boundaries of scale and real-world deployment.
Tsinghua Tianjic: first hybrid brain-inspired chip on a Nature cover
Tsinghua University's Tianjic chip demonstrated the world's first hybrid chip integrating both computer-science-based and neuroscience-based AI approaches on a single platform. The chip powered an autonomous bicycle that could self-balance, detect obstacles, and follow voice commands β all processed on one chip. Published as a Nature cover story.
Intel Hala Point: world's largest neuromorphic system (1.15 billion neurons)
Intel unveiled Hala Point, the world's largest neuromorphic system, built from 1,152 Loihi 2 chips with 1.15 billion artificial neurons and 128 billion synapses β roughly equivalent to an owl's brain. It achieves up to 20Γ faster processing and 100Γ better energy efficiency than conventional GPU-based architectures for certain AI workloads. Deployed at Sandia National Laboratories.
Nature Review: "Neuromorphic computing at scale"
A comprehensive Nature review paper assessed the state of neuromorphic computing, covering hardware advances (from memristive devices to large-scale spiking processor arrays), algorithm co-design, and emerging applications in edge AI, robotics, and scientific computing. The review concluded that neuromorphic systems are approaching the scale needed for practical deployment.
Memristive synapses achieve ultralow-energy brain-like learning
Researchers demonstrated HfOβ-based memristive synapses with asymmetrically extended conductance ranges, enabling ultralow-energy and highly stable artificial synaptic behavior. This addresses a key bottleneck for neuromorphic hardware: making individual "synapses" as energy-efficient and reliable as biological ones. Published in Science Advances.
Photonic computing uses photons (light) instead of electrons to perform calculations β especially matrix multiplications that dominate AI workloads. Light travels at, well, the speed of light, with virtually no heat. The key advantage: massive parallelism and orders-of-magnitude better energy efficiency. In April 2025, two photonic computing papers published in Nature on the same day signaled the field's coming-of-age moment.
Lightmatter Passage: photonic interconnect for chip-to-chip communication
Lightmatter (MIT spin-off) announced Passage, a photonic interconnect chip that uses light to link processors together β analogous to how fiber optics revolutionized long-distance data transfer, but at the chip level. This addresses the "interconnect bottleneck" that limits AI accelerator scaling.
Lightmatter Envise: first general-purpose photonic AI processor
Lightmatter unveiled Envise, a six-chip 3D-packaged photonic AI processor that performs matrix operations using light. The system demonstrated general-purpose AI inference comparable to electronic accelerators while consuming dramatically less energy per operation. Published in Nature.
Lightelligence (ζ¦ζΊη§ζ) PACE: large-scale opto-electronic hybrid computing
Chinese startup Lightelligence (ζ¦ζΊη§ζ) demonstrated PACE, an opto-electronic hybrid computing accelerator achieving 64Γ64 matrix-vector multiplications with industry-leading energy efficiency. Using photonic TSV (through-silicon via) packaging for tight opto-electronic integration, the system showed significant advantages in latency and power consumption. Published in Nature on the same day as Lightmatter's paper.
DNA and RNA computing harnesses the information-processing properties of nucleic acids. A single gram of DNA can theoretically store 215 petabytes (215 million GB) of data, and biological molecules can perform massively parallel computations. While still largely in the laboratory stage, breakthroughs in DNA data storage, programmable molecular circuits, and nucleic acid neural networks are pushing the field toward practical applications in diagnostics, data archival, and programmable therapeutics.
DNA data storage: from lab demos to startup-scale systems
Multiple companies (Catalog Technologies, DNA Script, Twist Bioscience) advanced enzymatic DNA synthesis for data storage, achieving faster write speeds and lower costs. Microsoft and University of Washington demonstrated a fully automated DNA storage system capable of encoding, storing, and retrieving digital data without human intervention. The field targets competitive cost parity with magnetic tape by ~2030.
Programmable DNA neural networks for molecular pattern recognition
Researchers built increasingly complex DNA-based neural networks that can classify molecular inputs β effectively performing pattern recognition using only nucleic acid chemistry in a test tube. These systems use DNA strand displacement reactions to implement weighted sums and thresholding, mimicking the basic operations of artificial neurons. Scale has grown from a few neurons to networks with dozens of computational nodes.
RNA-based smart therapeutics: computing inside living cells
RNA circuits capable of performing logic operations inside living cells are being developed for next-generation therapeutics. These programmable RNA "computers" can sense multiple biomarkers, perform Boolean logic, and trigger therapeutic responses only when specific conditions are met β enabling precision medicine at the cellular level. Multiple teams published advances in conditional gene regulation using RNA logic gates.
Beyond DNA/RNA, broader molecular computing uses chemical reactions, synthetic biology circuits, and even protein-based systems to process information. This includes chemical reaction networks that implement algorithms, synthetic gene circuits that function as biological computers, and molecular machines that perform mechanical computation. The field is deeply connected to synthetic biology and programmable matter.
Synthetic gene circuits: programmable cellular computers grow more complex
Synthetic biologists have built increasingly sophisticated genetic circuits in living cells, implementing multi-layered Boolean logic, memory, and even simple neural-network-like behavior. MIT and ETH ZΓΌrich teams demonstrated genetic circuits with dozens of logic gates operating reliably in mammalian cells β approaching the complexity needed for therapeutic applications.
Chemical reaction networks as universal computational substrates
Theoretical and experimental work demonstrated that chemical reaction networks (CRNs) are Turing-complete β meaning they can, in principle, compute anything a digital computer can. Researchers implemented CRN-based systems that solve optimization problems through molecular self-assembly and reaction kinetics, operating in massively parallel fashion.
Protein-based logic gates for in-vivo diagnostics
Researchers designed protein-based molecular circuits that can perform multi-input logic operations inside living organisms. Unlike DNA/RNA circuits, protein-based systems can interface directly with cellular signaling pathways, enabling "smart" diagnostics that detect disease biomarkers and produce visible or therapeutic outputs β bridging molecular computing with clinical applications.