Known-Good-Die Processing Benefits
Known-Good-Die Processing Contents
To first order, the probability of a system performing its function correctly is the cummulative probability of the components functioning correctly. The chart depicts the situation graphically. The probability of good devices in a lot that is shipped to a customer is plotted against the number of devices in the system (x axis) to predict the yield of the assembled module. This is fairly simplistic and ignores assembly defects and differing probabilites for different part types in an assembly, but it serves to illustrate the point. The expected yield of a system will be no better than the exponential noted in the figure below.
IC's that have been through the full manufacturing test flow have extremely low defect levels (probability of non-functional die or (1-Probability of Good Devices) in the figure above) when shipped to customers. The most realistic expectation for success in stacked SIP implementation is for each IC fabricator to perform all tests and reliability screens on the bare die prior to stack assembly. However, the requirement for thinning wafers for optimal height stacks has put added pressure on the wafer probing regime. There are KGD processes that are available in the industry.
Reliability
The other element of KGD processing is to improve the short term reliability that the customer sees by removing infant mortality devices from the lot prior to burn-in. The classic view of reliability is failures in time, portrayed on a "bath-tub" curve, which is a plot of failure rate vs time. It has been found empirically that many systems have a failure rate with the general form of a bathtub curve.
The maturity of the IC fabrication process can make a huge difference in the stratgy for reliability screening. Burn-in is a reliability screen that subjects the device to elavated temperature and voltage for an extended period of time (several hours duration) to cause weak devices to fail. The fails are subsequently detected using an operational test. Burn-in can be very effective in improving the reliabilty of a lot as shipped to a custormer, especially for new fab processes. There is diminishing returns for burn-in of mature lots of devices, with the end result that mature processes may achieve reliability goals without reliability screening of any sort.