P A P E R    T I T L E
A GaAs Fab's Approach to Design for Manufacturability

A U T H O R  /  C R E D I T S
Roger Garcia, Celicia Della and Subadra Varadarajan
Compound Semiconductor-1 Motorola, Inc.,
2100 E. Elliot Rd. MD EL720, Tempe, AZ 85284
Phone: 602-413-4834, email: rlyyj9@email.sps.mot.com

A B S T R A C T
Too often in manufacturing, one finds products which do not yield over all processes. Product engineering, test, design and wafer fabrication groups work after the fact to resolve yield issues. Products are redesigned, process windows changed or specification relief is asked for in an attempt to make up lack of proactive engineering. Through proactive engineering, products could be designed for manufacturability.

Another difficulty in design is the difference between DC tests parameters from the fab and RF results that must be taken into account in the design. MMIC devices are unique in the fact that RF parameters are required of them in all applications. At the same time, it is costly and difficult to perform 100% RF testing at the wafer level. It is much more efficient to perform DC tests as historically done in Silicon processing. DC tests can be done in two ways, i.e., Process Control Monitoring (PCM) and Unit Test whereby each product die is tested with a unique and usually shortened test scenario.

In order to help in the overall yield improvement, the fab can do several things to improve the manufacturability of a device. Through implant control and induced epitaxial variation, the process window can be artificially induced on a first lot. Evaluation of this device can result in correlation between PCM, Unit Test and final RF test at the package level. Test capability and system variation can be taken into account. We call this the process corners method.

In this paper, we will describe this method for correlation of DC tests to final RF tests. Using this method, correlations greater than r2=0.8 have been achieved. It is our contention that having such correlation allows the exemption of on-wafer RF testing. Furthermore, it will be shown that once specification limits are set using a dose split of process corners method, yield results at PCM and Unit Test track quite well with final RF tests done on packaged parts. Moreover, data from the process corners method can be used to generate design rules that take into account the process capability of the fab. We will discuss the fab's contribution to improvements in design and testing for manufacturability as well as similar results for variation in heteroepitaxial material.