THE COURSE / M10 MEMORY / M10.7
Testing the stack
After this lesson you can:
- Say how a stack is tested as it's built, and how a die buried in it is reached.
- Explain how a failed via is repaired after stacking.
- Explain why heat limits how tall a stack can grow.
Loading the lesson…
Sources
- K. Tran (SK hynix), "The Era of High Bandwidth Memory," Hot Chips 28 tutorial, August 2016. TSVs through each DRAM die over a base die with the PHY, the TSV area, the DFT test logic and direct-access balls; 1,024 data I/Os from the base die to the host through the interposer, up to 256 GB/s a stack (HBM2, 2 Gb/s a pin); micro-bumps 25 µm across at 55 µm pitch; a molded known-good stacked die, 0.72 mm tall (Gen2); the test flow: wafer test, wafer burn-in, stacking, a logic test through IEEE 1500, hot and cold tests and repair, speed test; after assembly, repair of cells and of TSVs and micro-bumps by lane remapping through IEEE 1500; thermal dummy bumps for heat.
- A. Meixner, "HBM Shifts Testing Left To Preserve AI Chip Yield," Semiconductor Engineering, 12 May 2026. Known-good stacks begin with known-good dies; a 12-high stack takes from 3 to 12 test insertions; defective TSVs are repaired at each; testing uses JEDEC's direct access or IEEE 1500 through a few pads or micro-bumps, and JEDEC leaves room for sacrificial test pads; HBM is nearly half the cost of AI chips, eight stacks around a typical GPU, so one bad stack is costly; power more than doubles from HBM3 to HBM4, up to 100 W a stack ahead.
- J. Liu, B. Jaiyen, R. Veras and O. Mutlu, "RAIDR: Retention-Aware Intelligent DRAM Refresh," ISCA 2012. A DRAM cell is a capacitor and an access transistor, connected to a bitline and controlled by a wordline; cells sharing a wordline form a row; each bank has a row of sense amplifiers, one per bitline; opening a row from bitlines precharged to VDD/2 shares each cell's charge with its bitline, and the sense amplifier drives the bitline fully to VDD or 0 V, recharging the cells; a row is refreshed by opening it; the refresh interval has stayed 64 ms for several generations; tREFI 7.8 µs below 85 °C, 3.9 µs from 85 to 95 °C; retention log-normal, set by the leakiest cell: in 32 GB, over 10¹¹ cells, about 30 can't wait twice as long and about 1,000 four times as long.
- Micron Technology (K. W. Marr), "Method, circuit and system for determining burn-in reliability from wafer level burn-in," US 6,943,575 B2 (filed 29 July 2002, granted 13 September 2005). The reliability curve's three regions: infant mortality, random failures and wearout; infant-mortality failures, though a small fraction of parts, are the largest contributor to early-life failure rates; burn-in stresses ICs with high temperature and raised supply voltage; conventional burn-in tests packaged parts.