|
Physically-Aware Diagnostic Resolution
John Porche |
Shawn Blanton |
A test-set, fault-equivalence (TSFE) class within an integrated circuit (IC) is defined here to be a group of logical faults that have identical fault-simulation responses for a given test set. The number of faults in a TSFE class is the conventional measure of diagnostic resolution, which is an indication of the ambiguity associated with precisely pinpointing the failed location within a faulty IC. Distinguishing faults in a TSFE class will create two or more smaller classes that have fewer members than before, thus improving diagnostic resolution. This reduces the time required for PFA (physical failure analysis), and increases the likelihood of identifying the reason for failure via PFA since the amount of IC area to be examined is now reduced. Reducing the amount of IC area that has to be inspected reduces cost and time, thus allowing the sample size that can be submitted for PFA to be increased. The success of PFA, and consequently the rate of yield learning, can be improved however if all TSFE classes can be partitioned so that the amount of IC area that must be examined, on average, is minimized.
A TSFE class can be partitioned using a test that distinguishes any pair of faults within the class. Normally, this is accomplished by choosing a pair of faults that, when distinguished, maximizes the conventional notion of diagnostic resolution. As shown in Fig. 1, this method can be used to improve diagnostic coverage (the ratio of the number of TSFE classes to the total number of faults), a single measure of the overall resolution of all TSFE classes on a chip. Instead, the objective of this project is to develop a physically-aware definition of resolution (PADR) whose aim is to increase the success of PFA. Targeting a PADR directly reduces the number of tests required to meet diagnostic goals. For example, Fig. 2 shows that few tests are needed to substantially reduce the wire length associated with TSFE classes when it is targeted directly, as opposed to indirectly when focusing on improving conventional diagnostic coverage.
 |
|
| Fig. 1: Relative to conventional fault coverage (FC), additional tests are needed for high diagnostic coverage (DC) for ITC99 circuit. |
Fig. 2: Targeting physical metrics instead of diagnostic coverage, achieves the PADR goals in fewer tests. |
|
