Overview
In late August 2026, China's AI x hard-tech sector hit five engineering milestones spanning chemistry, photonics, and quantum technology:
1. Chemistry: Hunan Institute of Science and Technology (Wan Zhongmin / Ren Shuangshuang team) published in *Carbon Energy* a Co,Fe@NCNF-900 bifunctional zinc-air battery catalyst — ΔE = 0.72 V, peak power density 393 mW/cm², specific capacity 736 mAh/gZn, stable cycling 420 hours at 10 mA/cm². 2. Photonic computing: Tsinghua University's Fang Lu group published in *Nature Sensors* the FLARE fully in-memory photonic computing architecture — 7,378 monolithically integrated photonic memory neurons, 7.45 s average long-term memory retention, 4 GHz short-term dynamics, 61.87 attojoules per operation. 3. Optical chips: Lanzhou University's Prof. Tian Yonghui team published in *Nature Communications* a reconfigurable optical arithmetic logic unit (ALU) chip — 20 Gbit/s, 528 Gbit/s² computing density, 12.36 fJ/bit. 4. Quantum LLM: On Aug 27, Beijing-based Zhongke Guoguang Quantum launched Xenomi, billed as the industry's first quantum-enhanced large language model. 5. Quantum hardware: On Aug 29, QuantumCTek (Guodun Quantum) obtained a patent for a quantum computing board/chassis enabling flexible, low-cost control-network reconfiguration.
Zinc-air battery catalyst: Co,Fe@NCNF-900
Zinc-air batteries are strong candidates for next-generation low-cost storage (high theoretical energy density, low cost, intrinsic safety), but air cathodes must drive both the oxygen reduction reaction (ORR, discharge) and oxygen evolution reaction (OER, charge), whose sluggish kinetics limit efficiency. Per the Sabatier principle, an ideal bifunctional catalyst needs intermediate binding strength — with the d-band center as the key descriptor.
The team's approach was to co-tune the electronic structure rather than swap materials, in three steps:
1. "Coordination–blending–porosity" strategy: ZnCoFe-ZIF, polyacrylonitrile (PAN), and cellulose acetate (CA) are co-electrospun into composite nanofiber membranes (ZnCoFe-ZIF@PAN-CA), then converted via two-step pyrolysis into porous metal–nitrogen–carbon nanofibers (Co,Fe@NCNF-900). Zinc sublimation and CA decomposition create porosity, improving mass transport. 2. XAFS + DFT electronic-structure analysis: Synchrotron XAFS shows Fe as single atoms and Co as atomic clusters coexisting on the carbon support. DFT indicates adjacent Co species redistribute Fe charge density and negatively shift the Fe d-band center, lowering ORR/OER rate-limiting barriers. 3. Dual-scenario validation: In liquid ZABs, the catalyst achieved ΔE = 0.72 V, 393 mW/cm² peak power density, 736 mAh/gZn, and 420 hours of stable cycling at 10 mA/cm²; flexible ZABs ran 31 hours of stable cycling.
A smaller ΔE (gap between ORR half-wave potential and OER potential at 10 mA/cm²) means one catalyst handles both discharge and charge efficiently — 0.72 V is first-tier among published works, removing the need for separate Pt/C (ORR) + RuO₂ (OER) catalysts.
Tsinghua FLARE: in-memory photonic computing
Existing optical neural networks still rely on external optoelectronic conversion and digital caching for inter-layer connections and weight reconfiguration. FLARE's answer: couple photonic cavities with electronic cavities so memory becomes intrinsic to the photonic neuron. Long-term memory uses programmable charge storage to hold network parameters; short-term memory uses transient optoelectronic modulation for inputs and intermediate activations.
Measured results:
- 7,378 photonic memory neurons monolithically integrated, with row-column addressing and parallel parameter writing — programmable array scale up ~2 orders of magnitude over prior methods
- 7.45 s average long-term memory retention
- 4 GHz short-term dynamic response
- 61.87 attojoules (6.187 × 10⁻¹⁷ J) per operation, claimed at least 3 orders of magnitude below comparable GPU inference
- Demonstrated on multi-layer fully in-memory optical neural networks for autonomous navigation
- 20 Gbit/s throughput
- 528 Gbit/s² computing density
- 12.36 fJ/bit energy efficiency
- d-band shift ≠ productization: From a 1 cm² lab electrode to mass production, batch-to-batch catalyst stability, electrode coating uniformity, and scale-up cost remain unsolved.
- 7,378 neurons ≠ datacenter cards: Packaging density, optical-electrical-optical interface bandwidth, and inter-chip interconnect loss gate the path from lab chip to accelerator card; an estimated 2–3 years of engineering remain.
- 20 Gbit/s ALU ≠ production optical CPU: Instruction sets, control flow, cache coherence, and OS support still separate an ALU from a CPU.
- Xenomi ≠ quantum advantage: Without public head-to-head benchmarks against purely classical LLMs, "quantum-enhanced" risks sliding into "quantum-washed."
- Patent ≠ product: Signal integrity, thermal design, and maintainability still stand between QuantumCTek's chassis patent and volume manufacturing.
- Science Net blog (Wan Zhongmin & Ren Shuangshuang team, Hunan Institute of Science and Technology), Aug 29 — *Carbon Energy*
- EEWorld / Electronic Engineering Album, Aug 29 — Tsinghua FLARE, *Nature Sensors*
- China Industry News via Toutiao, Aug 29 — Lanzhou University optical ALU, *Nature Communications*
- IT Home, Aug 29 — Xenomi quantum-enhanced LLM
- Photonic Box / QUANTUMCHINA, Aug 29 — QuantumCTek patent
- Ruicai Wang, Aug 29 — Guoguang Quantum multi-million-yuan strategic financing
Unlike an "optical accelerator" (matrix multiplication only), FLARE is positioned as an end-to-end optical computing stack: sensing, decision, and action can close the loop in the optical domain.
Lanzhou University reconfigurable optical ALU
Where FLARE targets optical neural networks, the Lanzhou optical ALU targets the "optical CPU" path — having light itself perform basic logic. The chip uses a reconfigurable bidirectional design executing adders, subtractors, and multi-operand comparators, at:
Unlike Lightmatter's optical-accelerator route, this chip aims at the foundational architecture of a photonic general-purpose CPU — the article likens it to the moment the 80387 coprocessor evolved into the 80386 general-purpose processor.
Quantum: Xenomi LLM + QuantumCTek patent
On Aug 27, Zhongke Guoguang Quantum launched the Xenomi LLM family, the first self-described quantum-enhanced LLM, integrating quantum techniques into decision-making and reasoning for scientific communication, routing, and agent decisions — deployed on a quantum computing intelligent experiment platform and the Kuami multi-agent system. On Aug 29, QuantumCTek secured a quantum computing board/chassis patent enabling flexible, low-cost control-network reconfiguration. Together they suggest AI moving into quantum (quantum algorithms inside an LLM) and quantum moving out (hardware deployable like servers).
The caveats
Questions for the next 6–12 months
1. Can Co,Fe@NCNF-900 hold ΔE = 0.72 V at 100-gram batch scale by 2027 H1? 2. Can FLARE scale single-card neuron counts toward 100,000 by 2027 H1? 3. Can the optical ALU execute a fuller instruction set (beyond add/subtract) by 2027 H1? 4. Can Xenomi publish public quantum-vs-classical comparison benchmarks by 2027 H1? 5. Can QuantumCTek's board/chassis enter small-batch pilot production by 2027 H1?