The additional risks in 3D packaging basically result from two facts:
1 . Stacking multiple dies into a 3D package also multiplies the probability of having a defective package in the end. In final test a package yield of 90 percent is realistic for standard packages. For a four-dies stacked package having 90 percent yield per die will bring down the overall yield to only 60 percent.
2. Beyond the risk of a bad die, there are additional components and process steps that may fail, e.g.:
- connections / bonding
In any case, bad elements will corrupt good ones – and finally the overall package.
Driven by the limitations of the established test point – wafer probing and final test – and the additional risks of 3D packaging, new test insertion points need to be considered. The additional test insertions must avoid good components that are added to bad stacks. The goal is to ensure “Known Good Stacks.” As a result, having in-process test points is particularly important before high value components (e.g. memory) are added. Smart decisions about the test insertion point and the test equipment will keep the cost for in-process test below the (opportunity) cost for non-testing.
The equipment for partial stacked die in-process test must fulfill economic as well as special technical requirements.
Technically, the challenge is to handle bare dies which are much more sensitive than packages. Economically, high parallel test is a must to reduce cost of test. Additionally, reusing equipment that has already been well-established will speed up the time-to-yield and reduces teething problems. Wafer probing equipment seems to be the best fit because of bare die handling requirements. However, wafer probers will not fully support the form factors of the rectangular substrate as they are designed for round wafers.
Final test equipment usually does not handle packages that are as sensitive as bare dies. Thus, special features need to be ensured:
The handler needs to accommodate the sensitivity of the bare dies in the partial stack. Physical support from the back-side of the substrate, respecting keep-out zones and avoiding unnecessary contacts during retest, are approaches to meet this. The contactors have to be able to satisfy yields even with low contacting force. This way the physical stress is reduced. High parallel test also requires sufficient probe compliance to accommodate for deflection.