Calculations

How to Calculate Yield, Throughput, and Test Capacity for Smart Home Devices

Work the five formulas that govern a smart home device line, rolled yield, flashing throughput, RF test capacity, molding scrap, and final test load, with real units and worked numbers.

A smart home device travels through five or six process gates before it ships: SMT assembly, firmware flashing, RF calibration and test, housing molding and final assembly, functional test, and pack out. Each gate has one governing calculation, and if you can run those five with real inputs you can predict output, staffing, and scrap for the whole line. This guide works each formula with numbers from a typical Wi-Fi smart plug program running 3,000 units per shift. Pull your inputs from MES yield logs, test station cycle logs, and molding shot counters, not from the quoted numbers in the original capacity plan.

Start with first pass yield: FPY = units passing on the first attempt divided by units started, computed per station. Rolled throughput yield multiplies the stations: RTY = FPY1 x FPY2 x FPY3, continued across the line. For the smart plug, SMT at 99.2 percent, AOI and rework verification at 98.5 percent, RF test at 97.8 percent, and final test at 99.5 percent gives RTY = 0.992 x 0.985 x 0.978 x 0.995 = 0.951, so start 3,155 boards to ship 3,000 units. The PCB Assembly Yield calculator also converts defect density to yield with yield = e^(-DPU); a 420 joint board at 100 DPMO gives DPU of 0.042 and predicts 95.9 percent.

Firmware flashing throughput is units per hour: UPH = fixture positions x 3600 divided by cycle seconds. Cycle time is image size over programmer write speed plus verify and handling. A 16 MB image over SWD at 1.2 MB/s writes in 13.3 seconds, CRC verify adds 6 seconds, a boot check adds 10 seconds, and load and unload adds 12 seconds, so 41.3 seconds total. A 4 position gang fixture yields 4 x 3600 / 41.3 = 348 UPH. The Firmware Flashing Throughput calculator lets you compare that against a 16 position bed of nails at a 60 second cycle, which delivers 960 UPH from a single operator.

Wireless test capacity converts RF cycle time into stations required: stations = required UPH divided by (station UPH x utilization). A combined Wi-Fi and BLE test measuring TX power, EVM, frequency error, and RX sensitivity across three channels runs about 75 seconds, plus 15 seconds of handling in a shielded box, so one station produces 3600 / 90 = 40 UPH. To feed 400 units per hour at 85 percent utilization you need 400 / (40 x 0.85) = 11.8, so 12 stations. The Wireless Test Capacity calculator also handles a 5 percent retest loop, which pushes demand to 420 UPH and the answer to 13 boxes.

Plastic housing scrap has two components: running rejects and startup scrap. Scrap rate = rejected parts divided by total parts molded, but cost it in material terms. A 38 g housing with a 6 g cold runner on a 4 cavity tool consumes 158 g of PC/ABS per shot at about 2.80 USD/kg, so each rejected shot burns 0.44 USD before machine time. Startup typically scraps 15 to 25 shots per changeover, which is 60 to 100 parts on 4 cavities. The Plastic Housing Scrap calculator totals both against shot counter data; three changeovers per day at 20 shots each adds 240 scrapped parts before a single quality reject.

Final functional test load starts from takt time: takt = net available seconds per shift divided by demand. A 7.5 hour net shift is 27,000 seconds, so 3,000 units needs a 9 second takt. If the functional test cycle, covering power on, relay actuation, Wi-Fi join, and a cloud handshake simulation, runs 120 seconds, stations = 120 / 9 = 13.3, rounded up to 14. Then add the retest stream: at 4 percent retest with a 150 second diagnostic cycle you need 3,000 x 0.04 x 150 / 27,000 = 0.67, so one more bench. The Final Functional Test Load calculator rounds these and flags any station planned above 90 percent utilization.

Chain the five results into one capacity check. RTY says start 3,155 boards; flashing at 348 UPH covers only 2,610 units in 7.5 hours, so flashing, not RF test, is the bottleneck in this example and needs a second fixture. Recompute monthly with trailing 4 week data and a minimum sample of 2,000 units per station, because a yield figure built on 200 units has a confidence interval wider than the difference you are trying to detect. Take cycle times from stopwatch studies of 25 or more cycles, not from equipment nameplates, which routinely overstate real throughput by 10 to 15 percent.

Published 2026-07-02.