Packaging Cost

Cost Per Unit in Advanced Packaging and Test: How to Build a Defensible Quote

A money-focused breakdown of advanced packaging and test cost per unit, from substrate and assembly to test seconds and yield-adjusted scrap.

In advanced packaging the substrate is usually the single largest bill of materials line. A high-density ABF substrate for a large flip chip package commonly runs 8 to 35 dollars depending on layer count and body size, while a mainstream laminate BGA substrate might be 0.40 to 2.50 dollars. Use Substrate Cost to price by layer count, body size, and yield loss, because a 4-2-4 buildup at 22 dollars versus a 6-2-6 at 34 dollars is a 12 dollar per unit swing before you touch assembly. Substrate scrap at 3 to 6 percent must be loaded back into the good-unit price.

Assembly cost is machine time plus consumables. A die attach placement at 1.5 seconds on a 65 dollar per hour machine rate is about 0.027 dollars of machine time, but a flip chip mass reflow with underfill and 240-wire bonding stacks up fast: figure 0.03 dollars die attach, 0.05 dollars for a 28 second wire bond cycle, plus underfill material at 0.08 to 0.20 dollars. Labor is a thin slice on automated lines, often 0.02 to 0.05 dollars per unit at one operator per three tools, so machine time and materials dominate, not headcount.

Test cost is priced in tester seconds, and this is where quotes leak money. Cost per second on a mainstream final tester runs 0.003 to 0.010 dollars once you load depreciation, handler, and floor. A 4.2 second program at 0.006 dollars per second per site, run 8-site parallel, is 4.2 times 0.006 divided by 8, or 0.00315 dollars per device. Wafer sort adds its own seconds. Package Test Cost and Final Test Throughput let you see how parallelism divides fixed tester cost; going from 8 to 16 sites can nearly halve the per-unit test dollars if the program supports it.

Burn-in is often mispriced because estimators forget oven-hour occupancy. If an oven costs 55 dollars per hour fully loaded and holds 3840 devices for a 26 hour cycle, that is 55 times 26 divided by 3840, or 0.372 dollars per device just for burn-in oven time, before board depreciation and yield fallout. Burn-in Capacity gives you occupancy so you can spread that oven cost correctly; halving devices per board doubles this number to 0.74 dollars, which is enough to make a marginal quote unprofitable.

Yield loss is a cost multiplier, not a footnote. If your fully loaded cost to a given step is 30 dollars and cumulative good-unit yield through that step is 94 percent, your effective cost per good unit is 30 divided by 0.94, or 31.91 dollars. Every point of yield below plan adds cost geometrically at the back end where the most value is embedded, so scrapping a fully assembled and tested module at 96 percent versus 99 percent yield can swing cost per good unit by 3 to 4 percent on a high-value part. Use Advanced Packaging Yield and Chiplet Assembly Yield to model this.

Overhead and equipment burden belong in the machine rate, not bolted on afterward. A pick-and-place or bonder rate should already carry depreciation over a 5 to 7 year life, maintenance at 6 to 10 percent of capital annually, cleanroom floor at 150 to 400 dollars per square foot per year, facilities gas and power, and utilization. A tool that costs 1.2 million dollars depreciated over 6 years is 200,000 dollars a year; at 6000 productive hours that is 33 dollars per hour before maintenance and floor, which is why realistic loaded rates land at 55 to 90 dollars per hour.

Build the quote bottom-up and stack per-unit lines: substrate, die attach, flip chip and underfill, wire bond, molding, test seconds at sort and final, burn-in oven hours, then divide the whole cost by cumulative yield to reach cost per good unit. Add a target margin of 12 to 25 percent depending on volume commitment and NRE recovery. Chiplet modules carry NRE for KGD screening and known good substrate that must amortize over program volume; a 400,000 dollar NRE over 2 million units is 0.20 dollars per unit that a spot-price quote will miss.

Estimates go wrong in predictable places. The three biggest are underestimating substrate escalation on high layer counts, quoting test at ideal parallelism the program cannot actually hit, and forgetting to convert every step cost to a good-unit basis. A quote that reads 42 dollars at nameplate cost is really 45 to 47 dollars per shippable unit once 92 to 94 percent cumulative yield is applied. Run Substrate Cost, Package Test Cost, and the yield tools together, and reprice whenever layer count, parallelism, or burn-in board loading changes, because each of those moves the number by tenths of a dollar to several dollars per unit.

Published 2026-07-01.