Diffusion Furnaces: How to optimize.

When optimizing diffusion furnaces, often your issues are quite simple. As an example, I was once working with a furnace that was running a very low temp anneal. The top of the furnace was overshooting. The vendor suggested an over-the-top, capital intensive redesign of the pedestal and boat. I had 2 suggestions. Increasing the N2 flow to dampen out the overshooting, and tuning the recipe to minimize it. We were able to eliminate the overshooting while maintaining throughput, and expended zero capital to do so.

Often the fixes revolve around finding little things that weren't properly optimized at tool installation, or perhaps where the person who created the proto-type process recipe made a minor mistake that is then propagated to every child recipe. If this has gone on for several years and has spawned many child recipes with the same flaw, this can involve large expenditure of resources, involving going to the PCRB, and then split lot and pilot lot to fix the issue. It is very important to scrutinize progenitor recipe closely and make sure that it has no flaws. Otherwise, you can find yourself expending much effort to fix a problem that started many years ago. It is definitely an example of something where an ounce of prevention equals a pound of cure.

Another classic problem that I have seen are incorrect transitions from spike to profile control, or from ratio control to 100% profile control. This often causes an over-reaction by the heater element, which causes a power spike, often to 100% power, which isn't good for either the heater element nor the product. Depending on which furnace vendor you are dealing with, sometimes this can be handled by setting maximum power limits. For others, it requires employing some recipe writing techniques that will prevent the furnace from having a violent reaction to transitioning from spike or ratio control to profile control during the deposition/growth step.

One of the easier problems to avoid is to employ Copy Exactly. On more than one occasion I have seen issues arise from someone playing Pablo Picasso with the furnace design or furnace model or furnace vendor. One extreme case that I saw a fab had purchased many diffusion furnaces from fire sales of fab closures. It required having duplicate inventories of spare parts for each vendor that they owned. The situation was so extreme that even when the furnace was from the same vendor and the same model number, because it had been purchased from fire sales from different fabs, and each fab had configured it to its own needs, unique maintenance solutions were required for furnaces that were next to each other that ran the same process because internally the equipment was different. The result was that while the initial purchase price had been a steal, the long term cost of ownership was extremely high. This cost included needing different spares sometimes for individual furnaces, as well as requiring documenting in specifications as to how to do maintenance on that piece of equipment, as well as needing to train the technicians on how to repair each variant of each furnace. The bottom line is that by not employing Copy Exactly, while it may be possible to get bargain basement prices on equipment purchases, by not having just one furnace model from one vendor the long term result is that duplicate spare parts inventories are required, as well as duplicating training efforts on each furnace vendor, and if you have different models from the same vendor, training on how to repair each model number.

When designing the recipes and manufacturing process, long term, it is cheaper and you get higher output if the recipes and processes are tuned towards quality over quantity. One sad instance that I observed quality, at best, was a secondary consideration, at least in recipe and process design. The by-product of this was the requirement for 100% inspection after every step, resulting in metrology being a bottle-neck, as well as a high scrap rate due to poorly designed processes and recipes. Better designed processes and recipes would have allowed for a standard sampling scheme rather than 100% inspection. By not designing in quality first, long term resources were expended filtering out the bad material. Any perceived productivity improvements that may have existed by designing recipes and processes that were tuned for throughput only disappeared when the high scrap rate and high costs of 100% inspections were considered.

These are just some examples. If you need an outside set of eyes to audit your diffusion processes, or would be interested in hiring a seasoned diffusion process engineer, please contact me!