Below are a few examples of projects I’ve worked on. This is not meant to represent the breadth of projects I’ve done, but does give some idea of the depth.
1 – Stage characterization: I was asked to characterize the mechanical performance of an integrated set of stages that provided micron precision motion in the X, Y, Z and theta directions. This was for a production application, and the goal was to determine the degree to which the design was capable of meeting the desired specifications. I first selected and purchased appropriate measuring equipment for the client: a laser interferometer, a capacitance probe and related instrumentation and electronics. Using LabVIEW, I integrated these instruments into a measurement system that also controlled the motion of the stages. I designed and had fabricated appropriate jigs and fixtures for the various measurements. The first step was to demonstrate the repeatability, accuracy and stability of the measuring system, which I measured to be on the order of 10 nanometers. I went on to provide data on performance of each component of the stage assembly, measured over different regions of travel, for different step sizes, velocities, and so on.
By collecting statistically significant sample sizes, I could examine error distributions to gain insights on the mechanical design. These integrated stages were large and fairly heavy, and I was able to show that the motion control system was doing a rather good job of closing the loop at the position encoders, but that there were significant discrepancies at the plane at which accurate stepping was required. This led to a lengthy inquiry through which I exposed the exact nature of the design shortcomings for each of the stages that made up the assembly. In particular, there were significant problems with the Z and theta stages, which I helped solve. To address the shortcomings in the X and Y stages, I proposed an alternate stage architecture which was ultimately adopted in the next generation product and has performed extremely well.
2 – Vibration Abatement: A piece of production test equipment used a large mechanical bridge to support a high power microscope looking down on a device-under-test (DUT). (By convention, the optical path is considered the Z axis.) The DUT sat upon a chuck that moved it on precision stages in the X and Y directions. These components were mounted to a large granite block that was suspended on vibration dampers to isolate it from floor vibrations that could otherwise create blurring of the visual image. The problem was that the image would move around for several seconds (swimming) after any movement of the X and Y stages. This not only created a poor impression of the equipment, but reduced throughput and created problems for algorithms that depended on machine vision through the microscope camera. My task was to eliminate this problem. This product was in the late stages of development, so the addition requirements were to move quickly and to change the design as little as possible.
The low frequency of the perceived swimming led me to the following hypothesis: Energy put into the system from the motion of the X and Y stages was exciting the natural frequency of the rocking mode of the system on the vibration dampers. This energy would be absorbed over a few seconds (by the dampers), but would meanwhile cause the microscope bridge structure to sway. I confirmed this by measuring with accelerometers and a capacitance probe.
The microscope was fairly heavy (~15 lbs.) and was mounted on a motorized stage (~50 lbs), all of which was suspended from the bridge which was roughly 1 meter tall. Despite the heavy structural members used on the bridge, I calculated that the significant inertial forces from rocking would lead to much more than the micron or 2 that could be tolerated. However, the displacements I actually measured were much greater than I expected. With more measurements I found that a bolted joint was flexing significantly, and I proposed a simple solution to reduce the overall deflections by more than 50%.
Since I knew that even an improved bridge was going to flex, I turned my attention to reducing the settling time. The bridge as designed was very heavy with a high center of gravity (CG), which made the job of damping vibrations harder. The simplest design change I could imagine was to grind pockets in the underside of the granite mounting block, thereby dropping the CG by the depth of the pockets. This turned out to be fairly quick and surprisingly economical, and the 2″ drop in CG improved the swimming performance another 25% or so – still not enough.
The existing vibration isolation system could not be further optimized, so I considered alternate suppliers. Settling time can be reduced at the expense of vibration isolation, so I needed to make sure that any new dampers still provided the necessary isolation performance. By this point, I had created a test to quantify the settling time, so now I devised a way to directly measure isolation performance over the desired frequency range (~3 t0 80Hz). (Both tests were implemented using LabVIEW to control the stages and a shaker, to pull in realtime transducer data, and to perform the necessary data manipulation.) I was able to find an alternate supplier with 3 to 4 times better settling performance with similar overall vibration isolation performance.
At this point, the performance of the system was acceptable, but I was asked to continue evaluating the overall design. It turned out to be easier to perform an experimental modal analysis rather than to do so through FEA simulations. Through this I found excessive deflections where the motorized microscope stage mounted to the stage, and a couple of other places. I proposed simple design changes that reducing the remaining vibration by about one third.
These changes were all implemented, and the product went on to perform (and sell) extremely well.
3 – Contact Resistance Characterization: This project was part of a larger R&D effort seeking to develop technology for the repeated electrical probing of large arrays of tiny metal contacts. These contacts were only about 10 microns across, and spaced only 40 to 50 microns apart. The primary goal for this (sub-) project was to determine the amount of force required for any given probing tip to make good electrical contact, with the additional variable that the contacts could be plated with one of several different metals (Au, Sn and Cu to start). This was for a high-volume testing application (millions of contact cycles), so another important goal was to determine how often it would be necessary to clean the tiny probe tips and what cleaning recipes would be effective.
My basic approach was to create a dedicated test cell that could accurately measure contact resistance over many thousands of cycles of single tips contacting variously plated substrates. This called for a great deal of repetitious testing and the manipulation of considerable data, so I assembled components that could be integrated through LabVIEW. A probe station was available that could hold silicon (or other) wafers on a chuck and drive it in the X, Y and Z axes. This was used to move a test wafer up to engage with a single tip, and also to step down, across a small amount, and back up so that the tip engaged with fresh, undisturbed contact metal each cycle. I specified and purchased (or designed and had fabricated) the other required elements: a precision milliohm meter, a miniature load cell and amplifier that could provide milligram level resolution, a data acquisition card, an assortment of wafers plated with the desired field metals, a simple switching matrix, and a miniature highly linear compliant element. The switching matrix was a simple breadboard with a few transistors to take the milliohm meter out of the circuit during connection and disconnection (to prevent micro arcing) while maintaining 4-wire resistance measurements as close as possible to the point of electrical contact. The compliant element was needed to obtain the desired load measuring sensitivity. (Without this, each vertical step of the chuck would result in a much greater increase in probing force than desired.)
This project demonstrated what I consider to be the excellent value of a good test set-up. There was a fairly significant investment up front to get the system to work as desired, and early on, virtually all the data taken was discarded. This is common with R&D where the most important parameters are simply unknown at the beginning. In this project, I uncovered issues of contamination, oxidation, poor correspondence between test substrates and real life contacts as well as other factors, not to mention some incredible subtleties of test technique. Having a good test setup made it possible to gather the data to work through those and then fairly easily acquire new, valid data once the problems were resolved. Ultimately this greatly facilitated the development of some groundbreaking results on a very promising new technology, published in the IEEE Proceedings. (See reference under “About Me.”)
4 – Glue joint optimization: The spindle motors on which hard disk drives are built must have exquisite precision. The width of the magnetic track on which the data is stored can be well under 2 um wide, so any radial perturbations in the rotation of such motors makes it harder for the head reading the data to follow that track. (Such perturbations are referred to as non-repetitive run-out, or NRRO.) To give an idea of the sensitivity, I would fail motors having an NRRO of fractions of a millionth of an inch at critical frequencies. To meet the required performance, the high-precision bearings used in such motors must be carefully preloaded. To do so, 3 of the 4 bearing races press against metal features in the hub and shaft (and are therefore rigid) and the 4th raceway, which can move freely with respect to the shaft, has the preload force applied to it. It is common practice to subsequently use adhesive between the shaft and inner race to lock in this preload force. After curing, the externally applied force can then be removed. The preload force in the bearings must remain stable over the life of the product, during which time it will be exposed to vibrations, mechanical shock, and – most notably – thermal shock.
I had final technical responsibility on one such project, and had spent a great deal of time and effort making sure optimal tolerances, surface finishes, cleaning techniques, and contamination prevention measures were in place, and was evaluating the performance of a few competing adhesives in the cleanroom where the pre-production line was installed. Two adhesives were of the anaerobic curing type (cyanoacrylate derivatives) that had the great advantage of reaching sufficient strength in seconds, making them amenable to high-volume production. The third was an epoxy that required several hours curing at elevated temperature (or longer at room temperature).
I selected 2 primary methods for evaluating adhesive performance. The first was the “ping” test: The motor hub served as a fixed mass connected to the motor base through the bearings, where the bearings behaved like springs. This created a simple harmonic resonant system where the frequency is proportional to the square root of the stiffness divided by the mass. Since the stiffness was directly related to the preload force, measuring this resonant frequency provided a quick and non-destructive way of evaluating the preload at any time. The second test was a destructive test in which the bearings were held fixed and the force required to press the shaft out of a given race was measured. (A third and important test was chemical out-gassing, not discussed here.)
I mostly obtained excellent results with the cyanoacrylates, but when looking at statistically significant sample sizes would always find at least one outlier – one result that made the numerical results no longer fit to a normal distribution. The advantages of these adhesives motivated me to improve the processes, and together with input from the adhesive applications engineers, I was able to achieve normal distributions (with small standard deviation) after fabrication. However, I could not achieve this after thermal cycling or accelerated life testing. The epoxy did not result in these issues, and so I looked at how to implement epoxy in high-volume production. With some creativity and iterations, the team developed simple and inexpensive fixtures that could maintain the preload force on a large batch of motors over the hours required for curing. While slightly more costly, the reduction in risk was well worth it, and I continued preparations for production using this process.
However, having collected a huge amount of data on the performance of motors built with cyanoacrylate based adhesives, I noticed a subtle but statistically significant difference in the performance of motors built with epoxy. This difference was lost in the measurement variability when looking at any single motor, but on an average basis the NRRO on these motors – perhaps the single most critical specification value – was definitely higher. At first I thought it might be a result of thermal distortions arising from the elevated curing temperature. Careful experimentation showed the same problem when curing at room temperature over a much longer time. (In fact, it looked like it might be slightly worse!) After some considerable probing through applications experts at the supplier it finally became clear that the process of polymerization does lead to internal stresses in the epoxy. These were clearly creating minuscule distortions in the bearing races. They suggested trying a ramp cure, in which the temperature rises very slowly over 2 hours, then holds for a short time before cooling. This eliminated the problem, and this production line was qualified by the customer shortly thereafter.