Samsung is ramping glass-carrier cleaning capacity from 20,000 wafers/month this year to 50,000 next year — a 2.5x jump that signals at least 2x more HBM4 and HBM4E production in 2025.
Glass carriers are temporary plates bonded under HBM DRAM wafers during thinning and drilling. Thinner dies = tighter warp control = higher stacks. HBM4 and HBM4E run 12-high and above, so carriers matter.
HBM4 on 1c DRAM + 4nm base die started shipping in February. Twelve-high HBM4E samples hit $NVDA and others in May.
Total HBM wafer starts are expected to rise ~40% YoY (180k/month → 250k/month). But HBM4-family mix is jumping from ~40% this year to ~80% next year once HBM4E scales.
So if carrier cleans are up 2.5x but total HBM wafers only up 40%, where's the extra glass going? Either into 12-high HBM4E volume ramp or into yield loss on the thinning step.
Either way, Samsung is betting big on next-gen HBM. More glass under the wafer = 12-high stacks without breaking the die.
Glass carriers are temporary plates bonded under HBM DRAM wafers during thinning and drilling. Thinner dies = tighter warp control = higher stacks. HBM4 and HBM4E run 12-high and above, so carriers matter.
HBM4 on 1c DRAM + 4nm base die started shipping in February. Twelve-high HBM4E samples hit $NVDA and others in May.
Total HBM wafer starts are expected to rise ~40% YoY (180k/month → 250k/month). But HBM4-family mix is jumping from ~40% this year to ~80% next year once HBM4E scales.
So if carrier cleans are up 2.5x but total HBM wafers only up 40%, where's the extra glass going? Either into 12-high HBM4E volume ramp or into yield loss on the thinning step.
Either way, Samsung is betting big on next-gen HBM. More glass under the wafer = 12-high stacks without breaking the die.
