Synopsys Yield Explorer binds design, test, and fab data to speed diagnosis

The problem of turning those little red dots on wafers into a meaningful idea of what is causing test failures has grown worse with every new process node. Once upon a time a visual inspection of the wafer usually did the trick. (OK, I really am old.) Today, it may take a combination of detailed knowledge of the critical-area and sensitivity data from the back-end design, all the insight that can be wrung out of the wafer-test data, and any suggestive hints that can be mined from the fab process data just to get some idea where to look.

The problem arises in that word combination. Design data, test data, and fab data all come not only in different formats, but from different engineering cultures. Combining them can mean putting experts on each type of data in a locked room and sending in sandwiches and coffee until they can sort out what each of them is seeing, share the knowledge, and reach a conclusion.

Synopsys, which has established positions in back-end design, test automation, and process modeling and control, is trying to automate this integration—beyond the level of automatically scheduling the sandwiches. Synopsys Yield Explorer, announced today at the International Symposium on Quality Electronic Design (ISQED) could be a big step in the right direction.

Yield Explorer is an ambitious combination of elements. The underlying foundation is a proprietary schema on an Oracle database, designed to absorb design, test, and manufacturing data. This foundation, as you might expect, runs on an enterprise-class server network.

The second major component is a GUI built around an interactive layout viewer. This choice of a viewing paradigm more or less automatically organizes all the data in terms of physical locations in the IC layout. The viewer is backed by a plethora of statistical and other analytical tools, all responding to conventional tcl through the GUI. The user can view a particular data set, or an overlay of several data sets, mapped onto the chip layout. Or the user can see conventional or custom statistical plots of a particular data selection. There is also, through the tcl interface, some capability for data-mining.

An important additional dimension to Yield Explorer is the ability to bring failure-analysis data, such as SEM images, into the database as well—in effect, picking a spot on the layout and looking at a SEM image of it.

The third, and least visible, portion of the package is where all the work gets done: in the applications that do the data analysis, statistics, mining, and so forth. These typically would reside on their own server clusters separate from the database, perhaps under the auspices of the particular organization that owns the relevant data. The applications, as I understand it, for the most part are those that already exist in these organizations. They are simply drawn together by the common database, the APIs, and the tcl scripts.

Synopsys product marketing manager Sagar Kekare says that the results of assembling all the data and viewing it through Yield Explorer can be dramatic. "We have seen problem isolation reduced from two or three weeks to two or three days," he claims. Kekare says for him the acid test of the product is that it is fast and accurate enough that a product or test engineer using the GUI can direct the failure-analysis team to the right node on the first try. That could save quite a lot of iteration.