Lessons in RF Manufacturing from a Chicago Sausage Factory

People often say RF is black magic and it sometimes feels that way. I remember one evening I was called down to the production floor to help troubleshoot a technical problem found during swing shift. There was a product going through final test and it would only pass if held at a certain angle. At first I was doubtful that this was the case, but I held it in my hands, watched the performance on the network analyzer, rotated the unit, and saw the performance degrade. First we suspected the VNA cables, but a golden unit was solid regardless of its orientation. Then we performed the standard “shake while listening for something rattling test” but couldn’t hear anything—plus the repeatability seemed to suggest it wasn’t due to FOD. X-ray imaging didn’t yield any clues. Eventually, we had to send it off to de-lid, found nothing wrong, and after real-seal the performance was stable. The best theory we had was that the problem was due to flux improperly cleaned from a feedthrough.

It was this type of problem that drew me to RF engineering in college. Circuits that only worked when you placed a finger in a certain spot. The gain reduced by the microscope light. While it felt like black magic we all knew that in reality it was physics too complicated to be fully modeled. To this day, I still find these problems fun until all of a sudden a revenue commitment is missed.

This fourth post in my series on applying commercial technology to the RF defense industry is focused on manufacturing. While successful product development requires innovation, modular architectures and miniaturization (the subjects of my previous posts), it is all for nought without effective manufacturing.

The production of RF and microwave hardware for the defense electronics industry relies on a level of artesian manufacturing rarely seen with high volume commercial products. Car factories can manufacture thousands of cars each week. Contract manufacturers can produce phones with extremely high first pass yield. What is it about their processes that allow for such high levels uniformity while often each RF component for a defense application is its own work of art?

Which brings us to a Chicago sausage factory in the 1970s. As many of us have experienced, business conditions at the factory required a move to a new facility. However, after the move, the sausages just weren’t as tasty. No one could figure out why. They tested the water in the new location and reviewed all the processes, but it wasn’t until a casual conversation with some employees was the solution found. At the old factory, it took about 30 minutes to walk from the cold storage to the smoke house. During this time the sausages warmed up. In the new factory, sequential processes were co-located and the sausages went in the smoker cold. By allowing the sausages to warm before entering the smoker, the new factory was able to match the quality of the old factory.

Improving First Pass Yield

How do we prevent these types of manufacturing issues? The first is to understand the two factors that contribute to their likelihood. The first is variation. Are the die always placed in exactly the same spot? How long are the bond wires? What is the tolerance on the capacitors? Are the MMICs from the same wafer? The second factor is the sensitivity to variation. What happens if the die is moved 5 mills towards the output? Will a small error in supply voltage cause an oscillation? Will the phase noise be impacted by that SSPA being tested on the next bench?

To optimize yield, manufacturing variation needs to be minimized and the product’s sensitivity to variation reduced. At Mercury Systems, we address both of these factors. Variation is reduced through automated processes such as the use of epoxy dispense and pick-and-place machines. For manual tasks, detailed instructions and thorough training ensure repeatability. This up-front investment in the assembly process results in a product that requires much less tuning. Once in test, automated test equipment ensures consistency and keeps process time low. In-process quality inspection finds any issues early in the process when they are easier to fix.

Sensitivity to variation is minimized during the design phase through advanced modeling and early collaboration between design and manufacturing. This step further improves yield by reducing the effects of the process variation that still remains. Mercury’s advanced manufacturing centers exemplify this commitment to efficient production. Maintaining low process variation while reducing the product’s sensitivity to variation enables high yield, a rapid production ramp-up, and helps keep costs low.