Bin and grade mix planner

How it works

Dies from one wafer do not all perform alike. Maximum frequency, power and leakage spread across the population, and the maker sorts the good die into bins that sell at different prices. This tool splits a distribution into bins, fills demand by price with down-binning, and compares two sets of cuts on revenue, average selling price and revenue per wafer.

1. Spread of the binning parameter. The parameter is described as a normal distribution (mean and standard deviation), a lognormal distribution (mean and standard deviation of the parameter itself, suited to leakage), or the measured values you paste. A lognormal is built from the ratio of spread to mean.normal: F(x) = Phi((x − mean) / sd); lognormal: F(x) = Phi((ln x − mu) / s), with s^2 = ln(1 + (sd / mean)^2) and mu = ln(mean) − s^2 / 2; measured: F(x) = share of values below x
2. Share of die in each bin. A bin takes the die whose parameter lies from its lower cut up to, but not including, its upper cut. A blank cut is open on that side. Where bins overlap, a die goes to the better grade. Die outside every bin are scrapped.bin share = F(upper cut) − F(lower cut), less the share already taken by a better bin
3. Units available. Good die per wafer times wafers gives the good die. An extra scrap share takes out fallout from other tests. The rest is split by the bin shares.units available in a bin = good die per wafer x wafers x (1 − extra scrap) x bin share
4. Fill demand, with down-binning. Bins are served in order of price, highest first. A bin sells its own die first. If demand is left and down-binning is allowed, it takes die from better bins, nearest grade first: a faster part sold as a slower bin, or a lower-power part sold as a higher-power one. You can allow none, one grade down, or any number.shipped = own die used + die taken from better bins; unmet = demand − shipped; excess = die in the bin not sold in any bin
5. Revenue, average price and revenue per wafer. Revenue is units shipped times the price of the bin they are sold as. Excess can be valued at a share of its bin price (salvage), which is zero by default. Average selling price is shown per unit shipped and per good die, which includes the die that stayed unsold.revenue = sum of shipped x price (plus excess x price x salvage share); ASP = shipment revenue / units shipped; revenue per wafer = revenue / wafers
6. Compare two cut plans, and test a process shift. The proposed plan is evaluated the same way and set beside the current one. The process shift table moves the whole distribution up or down by a share of a standard deviation, which shows how much a better or worse process is worth under each plan.shifted share = F(cut − shift); shift = number of sd x sd

Sources

The methods and default figures rest on these public sources. The wording here is our own; read the source for its full text.

• Speed Binning Aware Design Methodology to Improve Profit under Parameter Variations (opens in a new tab), A. Datta, S. Bhunia, J. H. Choi, S. Mukhopadhyay and K. Roy, Proceedings of the 2006 Asia and South Pacific Design Automation Conference, IEEE, 2006. A price-weighted yield over frequency bins, and the choice of bin boundaries to maximise profit.
• Evaluating the Effects of Cache Redundancy on Profit (opens in a new tab), A. Das, B. Ozisikyilmaz, S. Ozdemir, G. Memik, J. Zambreno and A. N. Choudhary, Proceedings of the 41st IEEE/ACM International Symposium on Microarchitecture, pp. 388-398, 2008. Chips of one batch are rated into frequency bins sold at different prices; the batch distribution decides profit. Cited for the binning setting, not for any figure.