Realizing Financial Payback from Yield Improvement

Laura Peters, Senior Editor -- Semiconductor International, 1/1/2000

Wafer probe yield improvements have a profound impact on a fab's profitability. For instance, a 1% increase in yield for a fab producing 25,000 wafers per month can result in an additional $6M a year. Nick Atchison and Ron Ross of TI's Silicon Systems (Santa Cruz, Calif.), use a commercially available yield analysissoftware package from Heuristic Physics Laboratories (San Jose, Calif.), Hyperview, to perform most of the analyses used in this series. Here we present methods for calculating return-on-investment (ROI) resulting from yield improvements.

To accurately quantify the financial impact of yield enhancement activities, random yield loss is separated from systematic yield loss using cluster analysis and critical area analysis (CAA), techniques automated by the Hyperview package. Using the negative binomial equation:

Y_i = Y_s

(1- AD)^a

and plotting mean probe yield as a function of die size for various products, cluster analysis determines the portion of yield loss from systematic problems (design, process or test) relative to defect density. CAA shows, by process layer, the probability that defects over a given size range will cause shorts, opens, missing contacts, etc. CAA also reveals any increased sensitivity to defects by comparing the CAA graphs for low-yielding product to those for product used for the cluster analysis. If CAA graphs of low-yielding product are comparable to good product, the data indicate a systematic problem.

Baseline yield improvements require knowledge of the source of defects. Killer probability can be determined, for instance, using in-line inspection tools such as KLA's 23XX tools and the Hyperview software. Improved defect density has a greater effect on yield of large dice as opposed to smaller dice. Financial impact is calculated using:

DR = AR oDY_iV_i i

where DR is increased revenue, AR is average revenue per wafer, DY_i is increased yield for product i and V_i is the wafer out volume for product i. If the average selling price and cost of packaging and final testing costs are known, a more accurate revenue estimate results from using:

DP = Y_FToDY_iV_i(ASP_i - ASY_i - FT_i)

i

where Y_FT is the final test yield, ASP is the selling price, ASY_i is the assembly cost per unit and FT_i is the final test cost per unit.

To illustrate, imagine an improvement in defect density to 0.45 defects/cm² from 0.50 defects/cm² with a revenue per wafer of $2000, volume of 25,000 wafers/mo with three products of die size 0.12 cm² (5,000 w/mo), 0.25 cm² (15,000 w/mo) and 0.50 cm² (5,000 w/mo). Using equation 1 and assuming a cluster factor, a, of 2.5, the yield improvements for the three die sizes are 0.56%, 1.05% and 1.80%. Using equation 2 the revenue increase is $551,000/mo or $6.612M/year.

Revenue gain from faster yield learning is calculated using yield curves and ASP curves over time. The financial impact is:

DP = G _t1*^t₂ F(t)dt

where F(t) = [Y_F(t) - Y_S(t)]V(t)[ASP(t) - ASY(t) - FT(t)]Y_FT(t)

Y_F(t) is the yield for fast yield learning in dice/wafer, Y_S(t) is the yield for slow learning, V(t) is the volume as a function of time, G is the gross number of dice/wafer, ASY(t) is the assembly cost, FT(t) is the final test cost and Y_FT(t) is the final test yield. Assuming an assembly cost of $0.80/unit and test cost of $0.20/unit and a constant final test yield of 95% with 600 gross dice per wafer and 200 wafers/mo, in the first four months,

Y_F = (0.40 + 0.10t)

ASP(t) = $6.00 - 0.25t x t

By calculating each segment until 12 months when the yield of the two curves become equal:

DP = $1.929 M.

The financial impact of accurately predicting yields results from producing too little or too much product. Assuming a high-volume product (5,000 w/mo) and a yield forecast 3% below the actual, the number of extra wafers produced is:

N = 3 x 5000 x 0.03 = 450 wafers, 450 x 2000 = $900,000.

The yield analysis software investment quickly pays for itself. For example, a $1M investment that results in a defect density reduction from 0.5 to 0.45/cm² ($551,000/mo) is paid back over less than two months (($1,000,000/$551,000/mo) = 1.81 month) and ROI is ($6,612,000/$1,000,000) = 661%. •