4-0 The Superhighway Beyond the Brain — Why Does Computing Power Need a Good Mattress?
AI servers rely on IC substrates for 'Space Transformation' as chip traces are 50,000x finer than PCB traces, precluding direct solder. IC substrates funnel nano signals, enlarge pitch, convert to PCB solder balls—the first hub linking compute power externally.
3-4 Breaking the Memory Wall —— CXL Interconnect Technology and PIM Processing-in-Memory
To overcome server capacity limits, CXL brings memory's 'USB moment' via PCIe for pooling & expansion, cutting cloud costs & sparking control chip demand. To end the 'data movement tax,' PIM integrates compute units into DRAM, enabling in-situ, no-movement computation. These shatter the Von Neuma...
3-3 Data Skyscraper —— The Magic of 3D NAND and Controller Chips
3D NAND vertical stacking breaks capacity bottlenecks; Lam Research's high-aspect-ratio etching is critical. QLC, for AI storage, mitigates inherent flaws via controller algorithms. aiDAPTIV+ extends SSDs into computational units, reshaping the AI storage value chain.
3-2 Edge Warriors —— LPDDR5X, DDR5, and CAMM2 Revolution
Edge AI needs speed, power, thinness. LPDDR5X soldered for integrity, not expandable. LPCAMM2, 20-year laptop revolution, modularizes LPDDR into thin, high-bandwidth, power-efficient, serviceable board, AI PC standard. This helps Taiwan's UMT & connector makers shift to high-margin precision comp...
3-1 The Lifeblood of AI —— HBM, TSV, and SK Hynix's Resurgence
HBM vertically stacks DRAM via TSV to overcome GDDR bandwidth. SK Hynix's MR-MUF solved thermals, beating Samsung's TC-NCF for AI chip dominance. For HBM4's 16-layer limit, Wuxi Ball's hybrid bonding is standard. TSMC intervenes in bottom manufacturing, becoming ultimate co-leader in memory packa...
3-0 The Famine of Computing Power —— Overcoming the Desperate 'Memory Wall'
Logic chip power surges, memory lags, forming a fatal 'memory wall.' Massive AI models spark a bandwidth crisis: data movement power exceeds computation, yielding 'data transport tax' & computing famine. Industry initiates revolution via HBM stacking & CXL, igniting bandwidth war reshaping global...
2-7-3 The Battle of Approaches — Intel PowerVia vs. TSMC SPR
Backside power sparks an architectural gamble. Intel's PowerVia routes power to front M0 (debug-friendly, less scaling). TSMC's A16 SPR sends power direct to bottom S/D, seeking performance. Extreme alignment aside, frees front space for ultimate scaling, ensuring strongest PPA.
2-7-2 — Extreme Engineering — Wafer Backside Refinement
BPD relies on extreme wafer-flipping: temp carrier support; DISCO grinds 99% Si to micron thickness; IR alignment etches nano TSVs for front contacts; low-temp backside power layer deposition. This micro-refinement sparks huge opportunities for semi equipment makers.
2-7-1 Physical Predicament — When the Building's Elevators Are Overwhelmed
Traditional front-side power delivery crowds signals & power, causing crises: thick lines take >20% space, hindering miniaturization; current via ultra-thin wires causes severe IR Drop, degrading performance & heat. The sole solution is to move the power delivery network to the wafer backside, fu...