I. Core Application Scenarios: Key Support for Breakthroughs in AI Hardware Performance

ALD technology, with its atomic-level thin film regulation capability (thickness control accuracy ≤0.1nm), material compatibility (capable of depositing oxides, metals, sulfides, etc.), and three-dimensional conformal property (uniform coverage of complex structures), focuses on enhancing the performance, energy efficiency, and reliability of AI chips, sensors, and storage devices in the field of artificial intelligence. Specific scenarios include:

①1.AI chips: Breaking through the bottlenecks of computing power and energy efficiency

<s:1> Advanced process transistor optimization: In AI chips with 7nm and below processes (such as GPU, TPU, and FPGA), ALD is used to deposit High-K gate dielectric layers (such as HfO₂ and ZrO₂) and metal electrodes (such as TiN and TaN), solving the leakage problem of traditional SiO₂ dielectric layers and reducing the gate leakage current by 1-2 orders of magnitude. At the same time, increase the driving current (Ion) by 10% to 15%, enabling the chip's computing power density to exceed 100 TFLOPS/W, meeting the low power consumption requirements for training large language models (LLMS).

<s:1> 3D stacked chip interconnect: Conductive films such as Co and Cu are deposited on the inner walls of the vertical interconnection channels (TSV, through-silicon vias) in Chiplet packaging to achieve low-resistance connections at the nanoscale gap (contact resistance <10⁻⁸ Ω·cm²), thereby enhancing the data transmission rate between multiple chips (>1 Tbps). Adapt to the trend of "computing power clustering" of AI chips (such as the interconnection of 8 Gpus in NVIDIA DGX H100).

<s:1> Enhanced heat dissipation and reliability: A composite heat dissipation film of Al₂O₃/SiO₂ is deposited on the surface of AI chips, increasing the thermal conductivity to 2-3 W/(m·K), solving the problem of thermal runaway at high power density (>500 W/cm²). Or deposit a SiNx passivation layer to reduce the charge accumulation of the chip during high-frequency operation and extend the mean time between failures (MTBF> 100,000 hours).

②2. Storage and computing integrated devices: Enhance data processing efficiency

<s:1> Preparation of new storage media: For the resistive variable memory (RRAM) and phase change memory (PCM) of AI memory and computing integrated chips, functional layers (such as HfO₂ -based resistive layers and Ge₂Sb₂Te₅ phase change layers) are deposited. By atomic-level doping of ALD (such as introducing Ti and Al elements), the crystallization/amorphous properties of the materials are regulated to reduce the size of the memory cells to below 5nm. The storage density has been increased to 1 Tbit/in², meeting the real-time access requirements of AI models for massive parameters (such as models with hundreds of billions of parameters).

Neural morphic computing hardware Ultrathin metal oxide films (such as TiO₂, ZnO) are deposited on the surface of artificial synaptic devices (such as memristors) that simulate brain-like chips. The defect density of the films is precisely controlled by ALD to achieve continuous adjustable synaptic weights (with an accuracy of 1%), simulating the learning and memory functions of biological neurons. Reduce the energy consumption of the AI inference process (1-2 orders of magnitude lower than that of digital chips).

③3.AI Sensors: Enhance environmental perception and data collection capabilities

"Miniaturized optical sensor: Deposition of ALD anti-reflection films (such as SiO₂/TiO₂ multilayer film systems) on the micro-lens surface of the CMOS image sensor (CIS) in the AI vision system can increase the light absorption efficiency by 15% to 20%, and improve the signal-to-noise ratio (SNR) of the sensor in low-light environments by 30%, meeting the high-precision image recognition requirements of scenarios such as autonomous driving and security monitoring.

<s:1> Multimodal sensor integration: Deposit functional films (such as Al₂O₃ moisture-proof layers and SnO₂ gas-sensitive layers) for the micro-sensors (such as temperature, pressure, and gas sensors) of wearable AI devices (such as smart bracelets and health monitoring patches), and achieve the miniaturization and integration of sensor arrays (reducing the size to less than 1mm²) through ALD. At the same time, enhance its stability in complex environments (such as reducing humidity drift by 50%).

④4. Quantum Computing: Promoting the integration of AI and quantum technology

Youdaoplaceholder0 Qubit protection and control: Deposit insulating/superconducting films such as Al₂O₃ and NbN on the surface of superconducting qubits (such as Josephson junction) to reduce external electromagnetic interference and material loss, extend the quantum coherence time (T1) to 100-500 microseconds, and provide a hardware foundation for quantum AI algorithms (such as quantum machine learning); Or, by precisely controlling the size and spacing of quantum dots (such as InAs/GaAs quantum dots) through ALD, precise regulation of single-electron tunneling can be achieved, and prototype devices of quantum neural networks can be constructed.

Ii. Market Size and Growth Drivers

Currently, the application of ALD in the field of artificial intelligence is driven by advanced manufacturing processes of AI chips and storage devices. The global market size is expected to reach approximately 300 to 500 million US dollars in 2024, accounting for 8% to 10% of the total ALD equipment market. The future growth logic includes:

The demand for AI chip computing power has exploded: The global AI chip market size is expected to reach 115 billion US dollars by 2027 (CAGR 35%), with advanced processes of 7nm and below accounting for over 60%. As a key process technology (a single GPU requires 10 to 15 ALD processes), the equipment demand for ALD will grow in tandem with the expansion of chip production capacity (expected annual growth rate of 25% to 30%).

The implementation of in-memory computing technology: The energy efficiency bottleneck of the traditional "separation of storage and computing" architecture has driven the research and development of in-memory computing chips. The global market size of in-memory computing is expected to exceed 20 billion US dollars from 2025 to 2030. The penetration rate of ALD in new storage media such as RRAM and PCM will increase from the current 5% to 30%, driving the rapid growth of device demand.

The popularization of edge AI devices: The demand for low-power and miniaturized hardware in edge AI terminals such as wearable devices and smart homes has soared. ALD, through sensor integration and chip energy efficiency optimization (such as depositing low-power films for edge AI chips), supports the improvement of computing power in terminal devices (such as edge inference speed >10 TOPS). It is estimated that the related market size will exceed 1 billion US dollars by 2027.

Iii. Competitive Landscape and Industrial Chain Characteristics

<s:1> International manufacturers dominate the high-end market: Enterprises such as Applied Materials (AMAT) of the United States, Lam Research of Lam Research, and Picosun of Germany occupy more than 70% of the global AI chip ALD equipment market share. Their equipment can achieve atomic-level uniformity (thickness deviation <1%) and high capacity (processing >100 wafers per hour) of 12-inch wafers. Monopolize the advanced process production lines of international chip giants such as TSMC, Samsung and Intel.

Youdaoplaceholder6 Chinese manufacturers accelerate domestic substitution: Enterprises such as China Microelectronics Corporation, North Star Innovation, and Microchannel Nano have launched 12-inch semiconductor ALD devices, with performance approaching international standards (for instance, the thickness control accuracy of China Microelectronics Corporation's Primo ALD device is ± 0.1A, and the step coverage rate is over 95%). The price is only 60% to 70% of imported equipment and has entered the production lines of SMIC, Yangtze Memory Technologies Co., LTD. The domestic market share is expected to increase to 15%-20% in 2024. In the future, there is a vast space for substitution in the mid-to-low-end processes of AI chips (such as 28nm and above).

The deepening of industrial chain collaboration: Equipment manufacturers and AI chip design companies (such as NVIDIA and Huawei hisilicon), as well as material suppliers (such as precursor manufacturers Merck and Tosoh) jointly develop customized processes (such as the High-K metal gate ALD solution for Gpus), reducing the R&D cycle from two years to within one year and accelerating the pace of technological iteration.

Iv. Challenges and Breakthrough Directions

1. Technical barriers and cost pressure

Complexity of advanced manufacturing processes: AI chips are advancing towards 3nm and below manufacturing processes, requiring the deposition of high-purity and low-defect films (such as oxygen vacancy density <10¹⁸ cm⁻³ in HfO₂). ALD equipment needs to be equipped with in-situ monitoring systems (such as X-ray photoelectron spectroscopy XPS, elliptic polarization thickness gauges). The increase in technical complexity has led to a rise in equipment costs (the price of a single high-end ALD device exceeds 20 million US dollars).

o Material and process compatibility: New materials for in-memory computing devices (such as organic-inorganic hybrid films) have higher requirements for the reactivity and purity of ALD precursors (such as metal-organic compounds). It is necessary to develop dedicated precursors and low-temperature deposition processes (<200℃) to avoid material performance degradation caused by high temperatures.

2. Geopolitical and supply chain risks

The US export control on semiconductor equipment to China (such as restricting the export of ALD equipment for 14nm and below processes) may affect the supply of ALD equipment for advanced processes of AI chips in China. It is necessary to accelerate the research and development of domestic core components (such as molecular pumps and mass flow meters) and precursors, and build an independent and controllable supply chain (for example, the purity of domestic precursors has been raised to 99.9999%).

3. Competition in emerging technology routes

Potential alternative technologies such as carbon-based chips (such as graphene and carbon nanotubes) and photonic computing may pose a challenge to the ALD demand for traditional silicon-based AI chips. However, the application of ALD in surface modification of carbon-based materials (such as graphene doping and carbon nanotube coating) can still create new market space.

V. Future Trends and Market Forecasts

The performance of the equipment is continuously upgraded: The next-generation ALD equipment will develop in the direction of "atomic-level precise control + ultra-high production capacity", supporting the deposition of three-dimensional structures (such as cross-chip transistor GAA) for 2nm and below processes. It is expected that the deposition rate of the equipment will increase to over 100 A /min by 2027, while the defect density will be reduced to less than 0.1 per cm².

<s:1> Diversified application scenarios: Extending from AI chips to cutting-edge fields such as quantum AI and brain-computer interfaces. For instance, quantum dot light-emitting diodes (Qleds) can be prepared through ALD for use in the optical interconnection layer of optical computing chips, or biocompatible films (such as TiO₂) can be deposited for brain-computer interface electrodes to enhance signal transmission accuracy (signal-to-noise ratio >50 dB).

The market size is growing rapidly: It is expected that by 2030, the global market size of ALD in the field of artificial intelligence will reach 1.5 to 2 billion US dollars, with a compound annual growth rate of 25% to 30%. Among them, the growth rate of the Chinese market is leading (exceeding 30%). Domestic equipment has achieved full replacement in the manufacturing process of mid-to-low-end AI chips and is gradually penetrating into high-end processes.

Vi. Conclusion

ALD atomic layer deposition is a core supporting technology for the breakthrough of AI hardware performance. Its application demands in advanced AI chip manufacturing processes, in-memory computing devices, sensor integration and other fields are clear, and the market prospects are broad. In the short term, it is necessary to break through the technical barriers and cost pressure. In the long term, with the continuous explosion of AI computing power demand and the acceleration of domestic substitution, the market size of ALD equipment in the field of artificial intelligence will achieve rapid growth and become an important growth pole of the global ALD industry.