IDDQ failure diagnosis is here
When I first read about IDDQ test in the early 90s, it seemed that the claims were too good to be true. IDDQ is a test method that finds defects by monitoring the current flow through VDD while the IC is in the quiescent, or quiet, state. Test had always been fairly complex and the claims that IDDQ testing got massive defect detection with just a few patterns and no circuit output measurements other than current didn't seem possible. After reading about silicon results, I realized it was something special. Indeed, it was quickly adopted throughout the IC test industry.
IDDQ test is common at many companies and accounts for a significant proportion of defect detection. Many people speculated that the increasing quiescent leakage current with newer fabrication processes and higher levels of integration would make IDDQ obsolete. IDDQ technology has, however, managed to keep pace with the test demands of newer devices. It provides detection of many defects and often uniquely detects defects that are missed by other tests. There are faster IDDQ measurement devices and techniques such as delta IDDQ that have enabled IDDQ to continue to be important in many test environments. Because IDDQ can uniquely detect some defects, it is very important for products such as automotive parts to have the ability to diagnose why devices are failing IDDQ tests.
The one historical shortcoming of IDDQ is that there was no way to automatically diagnose and pinpoint the root cause of a failed IDDQ test. Traditional PFA (physical failure analysis) takes too long for any device in volume production or ramping up to volume production. Wouldn't you like to know what was causing IDDQ patterns to fail? Fortunately, that problem is now solved; devices that fail IDDQ tests can now be diagnosed.
There is usually enough distinction between the patterns with too much current and patterns with normal current to allow failing patterns to be considered as failing and have diagnosis applied. Unlike stuck-at, transition, or most other scan patterns, however, IDDQ has the current measurement as an observation point only. The other techniques literally have all scan cells—over one million of them for a large design—as observation points. Recently, a new automated IDDQ diagnosis method was introduced that can handle multiple defects. In published research, a 130 nm automotive device used this IDDQ method to successfully diagnose defects in two products. A single defect was reported 87% of the time in one product and 71% in the other. Because the IDDQ diagnosis was proven to work effectively and produce good resolution, it is expected to work well with DDYA (diagnosis driven yield analysis) for yield improvement. See figure below.
Current signature of a device with suspected multiple IDDQ defects before and after sorting. If spatially different leakage paths exist in the chip, they will activate different leakage currents. Diagnosis will stop at the second jump if the second failing cluster is caused by a different defect.