We now use quartz-based crystals and their oscillators as standard components in nearly every design. There are countless versions available with different sets of specifications ranging from fairly good to very impressive, and at low cost. Whether crystal is used in a frequency synthesizer, as the clock for a processor, or to help extract a timing from a serial data stream, there are plenty of choices out there.
Of course, it was not always this way. I just finished reading a fascinating book “Crystal Clear: The Struggle for Reliable Communications Technology in World War II” by Richard J. Thompson, Jr (Figure 1). It’s yet another reminder of how much our industry has changed and advanced, and that these advances don’t just happen.
Figure 1 The book covers the little-known but huge and important story of the evolution of channel-tuning quartz crystals from a small-time “cottage” component to a high-volume, mass-production component at the heart of oscillator spurred by WWII. Source: Wiley-Interscience
It’s hard to believe that prior to WWII, only a few thousand crystals for radio tuning were fabricated each month with the vast majority being hand-made in small batches, mostly by amateur radio enthusiasts (hams). The electronics infantry had not yet come to fully understand what tuning control could do, since such crystals were unavailable, expensive, temperamental, and a challenge to get operating and keep somewhat stable.
Instead, most tuning was done using a master oscillator/power amplifier (MOPA) architecture with a passive LC-based oscillator followed by an amplifier/buffer, all tube-based, of course. It required at least 20 to 30 minutes of warm-up to come to a stable operating point. Due to technology advances and the needs of ships, especially aircraft, wireless was just beginning to assume a larger communications role compared to wireline links that Army used in the field.
Crystals promised better tuning—resulting in better accuracy and stability—and tighter channel spacing. However, the physics of the quartz resonator were poorly understood while the realities of mass-production were not understood at all. As WWII approached and then broke out, there was a legitimate discussion at all levels on whether wireless and radios should stay with MOPA or go crystal. The former was known and understood despite its limitations, while the latter had many technical, production, and field-experience unknowns (Figure 2).
Figure 2 Crystals in the early days were relatively large compared to modern versions; the typical pin spacing ranged from 0.5 to 0.8 inches. Source: http://www.af4k.com
Remember, this was before the inception of synthesized tuning, so engineers needed one crystal for each channel or frequency of interest, and frequencies were often under 10 MHz, or 10 megacycles per second, as they were called in those days.
Another complication in the potential shift to crystals were uncertainties in supply of the quantities of raw quartz needed, as our now-standard technology for growing these crystals as ingots didn’t yet exist. The only known source of natural quartz in quantity was from mines in Brazil, and it had imperfections and irregularities that compromised performance.
Jump to the end of the story: by a combination of major efforts in the lab, via trial and error in production using many large and small-scale manufacturers and, in the field, crystal production reached millions of units per month by the end of the war. This didn’t just happen by turning a crank faster. As the author notes “ultimately, the development of quartz oscillators became the second largest scientific undertaking in WWII after the Manhattan Project,” which created the atomic bomb.
It’s hard for us to envision the difficulties faced in those days, as frequency counters simply didn’t exist. The only way to verify the frequency of a ground and packaged crystal was to compare it to better standard using the horizontal and vertical channels of crude oscilloscope and a visual Lissajous pattern (Figure 3). While the “nulling” approach is a time-honored and viable way to measure an unknown versus a better standard—NIST still uses it for ultra-precision calibration in many cases—it requires considerable care and patience in equipment setup and implementation compared to using a high-performance frequency counter.
Figure 3 Lissajous curves can be used to compare frequency ratios—ranging from 1:1 to higher and lower—by comparing the number of horizontal and vertical lobes. Source: Electronics Club
One unwanted performance character of quartz crystals plagued the use of crystals until nearly the end of the war: aging. We are familiar with aging of components of various types in different applications due to a variety of electromechanical factors and types of stresses at the macro and micro levels.
The legendary analog-circuit designer and author Jim Willams cited it in one of his articles “This 30-ppm scale proves that analog designs aren’t dead yet“ and it’s still one of the finest examples of circuit-design excellence and elegance. To create a super-stable Zener-diode voltage reference, he ran tests on the actual diode he was using to establish the current at which it had lowest drift—and the current would be in a narrow range between 5 mA and 7 mA. Once he determined that current, he further aged the diode for 1,000 hours at that value.
The WWII crystal problem was this: crystals were tested and calibrated in production, but once in the field, they would drift over time. This, of course, negated their usefulness. Some of this drift was due to corrosion in the contact and holder as well as humidity, and those sources were addressed and rectified. But the aging still occurred, and teams of physicists, materials experts, crystal experts, and production-focused researchers studied the problem, offered various theories focused on issues related to the crystal surface and bulk, and pursued many false paths and hopes looking for a solution.
I’ll skip the details but eventually one thing became clear: there were really two stages of aging and thus possibly two aging mechanisms. First was an initial, short-term aging process which was completed over 24 to 48 hours, followed by a longer-term one which occurred over weeks and months. Washing the crystal surface, brushing to remove quartz micro-particles, and other cleaning steps “sort of” solved the problem, but only for a while; then the aging and drift reappeared.
However, as researchers worked with production experts, they began to more fully understand the physics of these crystalline structures and related microcrack development. In simplified terms, microparticles of quartz flaked off the crystal after grinding and cleaning, leading to microcracks—short-term aging. These microcracks then changed shape and form—long-term aging. The eventual solution to the problem was to use acid etching of the surface to reach the final crystal resonant frequency, which eliminated the micro-particles and resulted in a smooth, annealed-like surface at the microscopic level.
The result of this lengthy investigation into crystal aging and drift is typical of many “debug” scenarios. There are two symptoms which look, at first, as the two appearances of one problem, and thus there is only one problem to be solved. But you can’t be sure: perhaps there really are two different mechanisms? Will solving one problem also solve the other? Is there a correlation between the problems or just coincidence, perhaps with a common cause way, way back in the flow?
Have you ever had a hard-to-debug problem where the “single” observed problem really was a set of smaller problems which were somewhat related, or perhaps unrelated? Or have you had multiple smaller problems which appeared unrelated but which were eventually traced to a broader problem, such as a noise on a power rail?
- Good product, lousy crystal
- Functional Integration Changes Your Calibration Strategy, Part 1
- Functional Integration Changes Your Calibration Strategy, Part 2