August_EDFA_Digital
edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 20 NO. 3 6 where Y D = random defect yield, A = die area, D = defect density As an example, if an increase in the fab’s defect density causes a 3.5 point Y D loss toProduct 1, we can calculate the impact to Product 2, which has twice the circuit content as Product 1. For this example, assume: Prior to yield drop: Product 1 Y 01 = 97%, Y D1 = 70%. Assume Y 01 = Y 02 = Y 0 Using the Poisson model for the random yield component: • Product 1 yield pre Y D drop = Y 0 * Y D1 = 0.97 * 0.7 = 67.9% • Product 1 yield post Y D drop = 0.97 * (0.7-0.035) = 64.5% • Product 2 yield pre Y D drop = Y 0 * Y D1 n = 0.97 * 0.7 2 = 47.5% • Product 2 yield post Y D drop = 0.97 * (0.7-0.035) 2 = 42.9% So the overall yield impact to Product 1 is a 3.4 point degrade and a 4.6 point degrade to Product 2. Notice in this example, the systematic yieldcomponent didnot change. As a second example, insteadof a 3.5 point Y D loss, consider the impact of a 3.5 point systematic (Y S ) yield drop: • Product 1 pre Y S drop = Y 0 * Y D1 = 0.97 * 0.7 = 67.9% • Product 1 yield post Y S drop = (0.97-0.035) * 0.7 = 65.5% • Product 2 yield pre Y D drop = Y 0 * Y D1 n = 0.97 * 0.7 2 = 47.5% • Product 2 yield post Y D drop = (0.97-0.035) * 0.7 2 = 45.8% In this case, Product 1 experiences a 2.4 point degrade and Product 2 experiences a 1.7 point degrade. Because systematic yield does not scale with circuit content, the impact is proportionally larger on the better yielding (typically smaller) die. YIELD ACCOUNTING BY CIRCUIT TYPE A commonquestion that arises as part of yield analysis activities is whether an observed yield loss for a particular die subcomponent is as expected or not. As an example, consider the hypothetical 1cm 2 die illustrated in Fig. 4. Assume that the areaof each individual block of logic andmemory is known and that the electrical test allows the yield of each individual block to be broken out. A large number of die are tested and the average yield of each block is compiled. Table 1 contains the area of each block. Using the Poisson model, the expected yield of each block can be calculated. If the observed test yield of Logic Block 3 is 90% and the other blocks are yielding close to expecta- tions, that might suggest there is a systematic layout or test issue with that block. CONCLUSION In summary, this discussion provides a general over- view of yield model approaches. Ideally, the yield model takes into account relative circuit complexity and defect clustering. Thedierencebetweensystematicand random defect yield loss was discussed. Random defect yield scales with circuit complexity, while systematic yield loss generally does not. Both systematic and random defect components of yield lossmust be accounted for in a com- plete model. As the examples in this article demonstrate, with a basic understanding of common yield models, the reader can estimate expected yield losses, understand yield differences between designs, and even identify abnormal yield results within a particular design. Yield Fig. 4 Hypothetical chip floor planwith die area broken out by circuit type. Table 1 Expected yield by block for hypothetical die using a Poisson yield model A (cm 2 ) Y D (%) Expected * Y D (%) Actual Logic Block 1 0.4 89.4 Logic Block 2 0.2 94.6 Logic Block 3 0.1 97.2 90% Memory Block 1 0.2 94.6 Memory Block 2 0.1 97.2 Total Die 1.0 75.6 * Poisson D 0 = 0.28/cm 2
Made with FlippingBook
RkJQdWJsaXNoZXIy MjA4MTAy