February 11, 2013, anysilicon
Early on in Chip projects, yield is not taken very seriously. The common thinking goes – anyhow there isn’t much to do as this early point of time. However, there are actually several things you can do even before the Chip design starts, which will translate to clear savings.
1- Know your Yield
Yield has a great deal of impact only if production volume is high. If you plan to manufacture only a few tens of thousands of components, perhaps yield is not the most important topic in your project’s plan.
Yield can be roughly calculated or estimated before the project has even started. Yet, if you have calculate a yield target of 95% there is no reason to invest money and efforts to try improving the yield from (the calculated) 95% to 99% because that would not be possible. Therefore, it is important that you have calculated your yield and set that as a goal.
2- Consider Foundry Applicability
Semiconductor foundries are not taking any yield losses. It is not the fab responsibility whether your yield is high or low because they sell wafers and not dies. Therefore you should select the foundry the suits best to your Chip domain.
If you chip requires small node geometries go to GLOBALFOUNDRIES, TSMC etc. If you chip needs excellent RF performance go to: IBM, TowerJazz etc. The foundry can help you calculating the wafer yield based on their own process technology. If you can provide them with die size, number of layers, process node and options, they should be able to provide you with a very accurate yield figures for your project.
3- Match Design Team Experience to Your Project
If you have decided to outsource the frond-end and physical design activities to an external vendor, the main yield-related risk here is experience. If the design team does not have the relevant experience that matches your chip project (for instance: RF, High Voltage) you are really wasting your time. Don’t hire analog designer without high voltage experience if you need to design a 120V chip.
4- Select Silicon Proven IPs
More and more companies are shopping for Semiconductor IPs to help reduce time to market and minimize engineering cost. There are many IP vendors with high quality products and some with lower quality. The keyword here is risk minimization. You really want to make sure the IP blocks you are about to purchase and integrate into your chip are bug free and have been silicon proven and qualified for your process. Ask for test results and references.
[SWF] http://anysilicon.com/wp-content/uploads/2013/02/NRG-Technologies2.swf, 200, 250[/SWF]
5- Follow Package Design Rules
For simple QFN packages there are no real concerns besides following the assembly house design rules. However complex packages can reduce yield dramatically. If your chip uses a package that consists of a multilayer substrate with high speed signals, this substrate should be considered as part of the silicon die. Improper routing of high speed signals, for example, will make the substrate performance very marginal and thus result in failures during final test.
6 – Say No to Tight Test Limits, Say Yes to Better Hardware
The only place to measure yield is at the testing phase. And this is done by the ATE.
Great ASIC engineers often try to over-engineer the chip design and as a by-product also tighten up the test result criteria. These limits will have direct impact on your profit. Every device that fails to meet limits during the screening process will be scraped. Therefore, don’t create the perfect test specification. Make one that meets your system requirements.
Loadboards, sockets and probecards have different quality levels and therefore different cost. But since these are the actual physical interface between your chip and the tester, you want to make sure they have the right quality and durably to allow solid connectivity to the tester during the test period. Otherwise, lower quality hardware will shave off your yield figures. Sockets for example, have limited number of insertions; you therefore should buy a socket that meets your chip production volume. Bottom-line — don’t compromise on the quality of the hardware interfacing your chip.
There is so much more to write on this topic, we promise to write more articles in the future. Stay tuned.