Liu Yili's Perspective: The Edge Border of Silicon - From the US Military's 'AI Kill Chain' to Music Brand Spatial Audio

At the forefront of global technology and geopolitics, whether it's the US military's 'AI kill chain' deployed in the Western Pacific, Tesla's self-driving cars speeding down the roads, or Sony Music Entertainment (SME) precisely delivering 8D surround music to listeners' headphones through streaming platforms, the core technology behind them all is the same—edge computing (Edge Computing). However, even though it's the same edge computing, the 'survival environment' and 'computing logic' for military and civilian use are vastly different. This is not only a choice of technical routes but also a strategic game involving latency, survival rates, and silicon supply chains.

I. Core Concept: Why Do We Need 'Edge Computing'? Traditional cloud computing (Cloud Computing) is like having the 'brain in the US and the limbs in Taiwan'—all data must be sent back to the backend data center through a long high-speed network line, and then sent back after the calculation is completed. However, in the real world, this model has two fatal flaws: latency (Latency) and bandwidth bottlenecks.

On the World Cup field, the referee must make an offside decision within 0.5 seconds. If the image is sent to the American server and then sent back after the calculation is completed, the player would have already run to the halfway line. In the field of autonomous driving, a vehicle traveling at 100 kilometers per hour would have to wait for the cloud response if it encounters a sudden situation. This 50-millisecond delay is the difference between life and death. In high-risk modern battlefields, there's no need to mention it. The enemy's strong electronic interference will directly cut off this 'network line'. The core logic of edge computing is to 'directly distribute the brain's intelligence (computing power) to the limbs (terminal devices)'. Let the data be computed and decided immediately on the device where it is generated, at the moment it is generated.

II. Military vs. Civilian: Operating Principles and Tactical Logic 1. US Military Systems: Distributed Sensing, Edge-Based Decision Making Taking the US military's 'Joint All-Domain Command and Control (JADC2)' and Project Linchpin as examples, the drones, Aegis ships, and low-orbit satellites deployed in the first island chain are all independent 'edge nodes'. When a drone captures the tiny ripples of a submarine on the sea surface, it doesn't need to send several GB of high-definition images back to the headquarters in Hawaii. The small AI model embedded inside the drone's chip will directly filter the images on-site (Edge) and only extract a few KB of 'key coordinates and ballistic parameters', synchronizing them with various missile positions through tactical networks in seconds. The tactical purpose is to achieve extreme 'anti-interference' and 'survival rate'. Even if the rear command post is destroyed by missiles, the drones and Aegis ships on the front line can still independently complete the OODA loop (Observe-Confirm-Decide-Strike) relying on the edge chips on their bodies, forming a localized automated kill chain. 2. Civilian Technology: Ultimate Experience, Cost Control, and Privacy Compared to military scenarios, civilian edge computing pursues 'user experience, cost control, and data privacy'. In autonomous driving systems like Tesla (Tesla FSD), cameras around the vehicle generate massive amounts of images every second. The onboard edge AI chip must recognize 'whether it's a pedestrian or a cardboard box' within a few milliseconds and directly control the brakes. It cannot rely on 5G networks because entering a tunnel and losing network connection means losing control. In music and entertainment streaming, when a user listens to Sony's latest spatial audio (Spatial Audio) through their phone, the cloud only sends encrypted audio files. The phone chip (such as Apple A series or Qualcomm Snapdragon) uses edge computing to perform real-time spatial algorithm decoding on the local end, simulating the 3D sound field of instruments and the lead vocalist. This not only saves the enormous bandwidth cost of cloud servers but also provides an immersive experience without delay.

III. Differences in Chip Demand: Steel Hard Man vs. Top-Class Sports Car Since their ultimate goals are completely different, the physical hardware architecture required for their realization has gone in completely different directions. This is also why, although both are extremely dependent on TSMC, the technical specifications they require are different. 1. Core Process and Evolution • Civilian Edge Chips: Pursuing top-tier advanced processes (such as TSMC's 3nm, 2nm first-wave customers Apple and Nvidia), striving to pack as many transistors as possible into an extremely small volume. • Military Edge Chips: Not blindly pursuing the highest process, commonly seen are 7nm to 14nm mature or semi-advanced nodes, with some advanced systems adopting 3nm/5nm small chip (Chiplet) heterogeneous integration design, with stability as the first consideration. 2. Survival Environment and Tolerance • Civilian Edge Chips: Operating environment is relatively comfortable. Although autonomous driving chips need to pass automotive certification (AEC-Q100), they essentially operate in environments with air conditioning or cooling fans. • Military Edge Chips: The environment is extremely harsh and must have anti-radiation, high-temperature tolerance, and anti-shock capabilities. Chips must ensure they do not burn out or experience bit flips (Bit Flip) under the enormous G-force of missile launches, high-energy space radiation interference, and extreme temperature differences from -50°C to 150°C. 3. Computing Architecture Design • Civilian Edge Chips: Adopt fixed architecture (ASIC / NPU). The circuit architecture is fixed at the time of shipment, pursuing the ultimate 'power efficiency (computing power per watt)' to prevent vehicles from overconsuming power or smartphones from overheating. • Military Edge Chips: Adopt heterogeneous integration (FPGA + ASIC). The electromagnetic interference signals on the battlefield change instantaneously, so chips must have 'field-programmable (FPGA)' capability. If the enemy suddenly adopts new frequency band interference, the aircraft can directly refresh the chip circuit architecture through software during flight to immediately decode it. 4. Product Lifecycle • Civilian Edge Chips: Lifecycle is only 1 to 3 years. Consumer electronics products change generations once a year, and the iteration speed of chips is extremely fast. • Military Edge Chips: Lifecycle is as long as 10 to 20 years. Defense systems must ensure that the chip can still obtain the same spare parts and maintenance support ten years later.

IV. The Strategic Blind Spot Behind the Silicon Shield and the Ultimate Test of American Military Hegemony Whether it is the 'anti-radiation, high toughness, and CoWoS heterogeneous packaging' pursued by military chips, or the 'ultimate transistor density and low power consumption' pursued by civilian chips, these two seemingly parallel development tracks ultimately point to the same destination for crystal forging—Taiwan's TSMC. This precisely reveals the greatest and most fatal strategic paradox that the US military is currently facing in the Western Pacific: the 'AI kill chain' that the US military has built with all its national strength, the JADC2 global command system that it is so proud of, is essentially built on a supply chain gap that it cannot fully control. In recent years, the US military has been vigorously promoting the 'Chip Act' and 'localization of defense supply chains'. On the surface, it is geopolitical defense, but in essence, it is a collective anxiety about the decline of its own industrial strength. The US military is well aware that if a conflict breaks out in the Taiwan Strait, once the advanced packaging and FPGA edge chip production capacity on this East Asian volcanic island is cut off, the defense giants on the US mainland, no matter how much steel and explosives they have, can only produce 'World War II-level scrap iron' without a brain. At the same time, from Tesla's autonomous driving to the global digital entertainment empire of Sony Music, they will also collapse due to the cessation of the silicon pulse of consumer electronics. Thus, it can be seen that edge computing is no longer just a computing game for tech giants. It is the last digital neural network for the Pentagon to maintain US military global hegemony. This edge line woven by micrometers and nanometers of silicon is both the strongest blade of the American military machine and its most vulnerable Achilles' heel in the geopolitical game. The fate of this silicon shield will directly determine the direction of the tilt of the global power structure in the next century. V. Conclusion: TSMC's Hidden Dominance Between the Two Whether it is the 'anti-radiation, high toughness, and heterogeneous packaging (CoWos)' pursued by military chips, or the 'ultimate transistor density and low power consumption' pursued by civilian chips, the manufacturing destination is TSMC. The reason why the US military is vigorously promoting the 'Chip Act' is that they realize that if the first island chain goes to war, the FPGA edge chips used by the US military on drones to resist interference, once the supply chain from Taiwan is cut off, the arsenals on the US mainland, no matter how much steel they have, can only produce 'World War II-level' scrap iron without a brain. This strategic line woven by edge computing is the 'digital barrier' that cannot be crossed in current geopolitics. *The author works at WZMP LLP Legal Counsel in New York.