The first KILN prototype failed its calibration on a Thursday, passed it on Friday, and failed again the following Tuesday. Three engineers stared at the log for two days before someone said the sentence that ended the meeting: the filter is better when it is hot. Every failure happened in the afternoon. Every pass happened at 8 a.m. We had spent forty-eight hours debugging a feature.
This is the note about what happened next, which involved a thermal camera, one blowtorch, a written safety review, and a drive stage that now heats itself on purpose. It is the module we get the most mail about, and most of the mail is the same question: is heating a transistor on purpose not just making the defect a feature? Yes. That is exactly what it is, and the interesting question is when that is honest engineering and when it is marketing with a soldering iron.
What heat does to a drive stage
A transistor's behavior moves with temperature. In a measurement circuit, that movement is drift, the enemy, the thing the whole company exists to fight. In a drive circuit, the operating point shifting with heat means the saturation curve softens as the stage warms. Signals that clipped hard at cold clip rounder when hot. The harmonic structure changes. The module sounds, in the oldest word in the gear vocabulary, warmer, and for once the word has a mechanism you can put on a chart.
The afternoon failures were this: the calibration rig measured the drive stage at 9 a.m. and again at 2 p.m., and the response curve between those two passes moved by an amount that would have failed any output stage QC. It was not the oscillator, not the filter core, not the jacks. It was the drive transistor sitting four millimeters from a voltage regulator that idled hot. Proximity, not physics, was cooking the part.
The decision meeting, summarized
Four options were on the whiteboard. One: insulate the drive stage and ship a filter that measures identically all day. Two: heat sink the regulator and accept a 60% improvement in drift. Three: move the regulator and add thermal mass, cost unknown. Four: lean in, control the heat, and write a spec for a curve that moves.
We picked four, and the reason was measurement, not romance. When we drove the stage hard at operating temperature, the response was more consistent unit to unit than the cold measurements had been, because the thermal operating point stops depending on your room once you force it. A defect that varies with the weather became a specification that does not. The blowtorch session proved the boundary: with the flame at fifteen centimeters, the curve moved exactly as the model predicted, which meant the model was good and the part was understood.
What the production part does
The shipped drive stage holds its own temperature: a resistor network sized to warm the transistor to its operating point in four minutes and hold it there within one degree for the rest of the day. The panel carries a small warm-up LED that goes from amber to steady when the stage is in spec, which is the only status light on any OSCILLA module that exists to tell you when the instrument is ready to be imperfect. Warm-up takes four minutes. The manual says to make coffee. The manual is right.
The spec sheet prints both curves, cold and warm, and the batch certificate records the warm curve for your unit. What it does not print is a "warm" switch, because the heat is not a feature you toggle. It is the operating condition, the same way a woodwind player's reed has a temperature. You do not toggle a reed.
The honest-versus-marketing line
Here is the test we apply now, written down so it survives us: heating a part on purpose is honest engineering when the heated behavior is measured, specified, repeatable, and reproducible for the customer on the bench at home. It is marketing when the behavior is described in adjectives and defended with the word "mojo." KILN passes the first test. The one competitor module we measured that heats its drive with no published curve, no tolerance, and no reproducibility instruction fails it, and we told them so at a trade show, which is why we are no longer invited to that booth.
Corrections. An earlier draft said the regulator sat "two millimeters" from the drive transistor. Thermal imaging logs put it at four. The conclusion is unchanged and slightly less dramatic, which is usually how corrections go.
Filed under: drive stages, thermal design, KILN. Requests for the thermal model spreadsheet go to the bench address; we send it with the column headers explained and no warranty, because a spreadsheet about heat is a starting point for an argument, not an answer.