Grown Junction Transistor Process Flow
# Grown-Junction Transistor Device Process Flow: Melt-Dosed Czochralski Pulling & Pull-Rate Base-Width Control
The grown-junction transistor froze its own base width into the crystal as it grew. Built by Gordon Teal and Morgan Sparks at Bell Labs in 1951 — the first junction transistor of any kind, predating the alloy-junction process by a year — it set emitter and collector junctions not with a separate furnace step afterward, but by deliberately changing the dopant fed into a single Czochralski melt at two precise instants during one continuous crystal pull. A single growing germanium crystal solidified n-type, then p-type, then n-type again, in that order, along its own growth axis — so that slicing the finished ingot crosswise, at the right position, cut straight through a complete n-p-n sandwich that had never been touched by a human hand, a whisker, or a second thermal step. Base width was not measured after the fact; it was scheduled in advance as a pull-length-per-unit-time calculation, then simply let happen as the crystal grew.
---
## 1. Why a Single Melt Can Grow Three Doping Regions in a Row
A Czochralski puller holds a rotating seed crystal just touching the surface of a molten charge, slowly withdrawing it as the melt freezes onto the seed's tip one atomic layer at a time. Ordinarily the melt's dopant species and concentration are fixed before the pull starts, so the whole ingot ends up uniformly doped. Teal and Sparks's innovation was to treat the melt's dopant as something that could be changed *mid-pull*: drop a pellet of a different, opposite-type dopant directly into the melt while the crystal is still growing, in sufficient concentration to overwhelm whatever dopant was there before, and the crystal's growth front — still advancing at the same pull rate — starts incorporating the new dopant into every atomic layer it freezes from that instant forward. Do this twice, with the second pellet restoring the original dopant type, and a single pull produces an n-p-n sandwich frozen along the ingot's own growth axis.
---
## 2. Segregation, Overdosing, and Why the Transition Isn't Instantaneous
A dopant does not distribute itself identically between liquid melt and the solid crystal freezing out of it — the equilibrium segregation coefficient $k_0 = C_s/C_l$ describes how much more, or less, concentrated a dopant is in the freshly solidified crystal compared to the melt it just froze from. As the crystal continues pulling after a pellet drop, the melt's own composition is simultaneously evolving because the growing crystal preferentially rejects or absorbs dopant at the solid-liquid interface — a redistribution the Scheil equation describes for the bulk melt concentration as a function of the solidified fraction $f_s$:
Base width as a scheduled time-times-rate product. Because the crystal advances at a constant, independently set pull rate $v_p$, the physical distance between the two pellet-drop transitions is simply
where $t_1$ and $t_2$ are the pellet-drop timings a process engineer schedules against a stopwatch before the pull even begins. The transition itself is not perfectly abrupt — the new dopant must first overwhelm the residual concentration of the old one in the melt, a transient the Scheil relationship governs — but by using a large pellet charge relative to the melt's existing dopant inventory (the same *overdosing* logic the alloy-junction process used with oversized indium pellets to make depth saturation-limited rather than timing-limited), the transition region can be compressed to a small fraction of the total pull length, so that $W_B$ is dominated by the clean $t_2 - t_1$ time difference rather than by segregation-smeared uncertainty at either edge.
---
## 3. Why Pull-Rate Stability Is the Entire Game
Unlike the alloy-junction process, where base width is set by a furnace soak that can simply be repeated identically on the next boat if a batch comes out wrong, a grown-junction base width is locked in the instant the crystal finishes growing past the second pellet drop — there is no reworking a single ingot's doping sandwich after the fact. Every source of pull-rate variation — motor speed ripple, melt-level drop as the charge is consumed, operator reaction time triggering the pellet drop — propagates directly into base-width scatter, because $W_B$ is a product of a *time interval* and a *rate* that the same mechanical system is simultaneously trying to hold constant for the entire pull. This single-ingot, no-rework constraint is the central economic weakness of the grown-junction process relative to the alloy-junction process that superseded it within about a year: a bad alloy soak costs one furnace boat's worth of dice, while a bad grown-junction pull costs an entire ingot's yield, discovered only after growth, slicing, and test — far too late in the process to recover any of the material cost already invested.
---
## Grown-Junction vs. Alloy-Junction: Where Each Dimension Is Actually Set
| Process Decision | Grown-Junction (1951) | Alloy-Junction (1952) |
|---|---|---|
| When $W_B$ is fixed | During crystal growth, as $(t_2-t_1)\cdot v_p$ | After growth, in a separate furnace soak |
| Reworkable without re-growing? | No — one ingot, one outcome | Yes — re-alloy another wafer from the same boat |
| Dominant scatter source | Pull-rate and melt-level drift during the whole pull | Furnace timing drift (minimized by pellet saturation) |
| Transition sharpness controlled by | Pellet overdose vs. Scheil segregation smearing | Pellet volume vs. diffusion-limited regrowth depth |
| Batch unit | One ingot per pull | Many dice per furnace boat |
| Junction count per growth run | Exactly one n-p-n sandwich per ingot | As many junction pairs as dice loaded in the boat |
---
## Why the Grown-Junction Process Flow Was, in Turn, Superseded
Teal and Sparks's grown-junction transistor proved, a full year before the alloy process existed, that a furnace-controlled dimension could beat a hand-placed one — but the specific mechanism it used to control that dimension carried two structural weaknesses that the alloy process immediately corrected:
1. No post-growth correction. A pull that drifted off its scheduled pellet-drop timing produced a flawed ingot that could only be discovered after the entire multi-hour growth cycle finished, sliced, and reached electrical test — with no way to salvage the base width without discarding the crystal and starting over.
2. One doping sandwich per pull. Every ingot produced exactly one usable n-p-n region no matter how long or wide the pull was grown, while the alloy process could load dozens of individually processable dice into a single furnace boat and alloy them all in one soak.
Within about a year, the alloy-junction process took over as the commercially dominant route precisely by decoupling junction-forming from crystal growth — the same furnace-recipe philosophy the grown-junction transistor had introduced, applied to a step that could actually be repeated, corrected, and batched independently of how the underlying wafer material had been grown.
Read the grown-junction transistor through a *scheduled-in-advance* lens rather than a *just another transistor* lens: it was the first device whose base width was a number written down on a process sheet before any crystal touched a melt, computed from a pull rate and a stopwatch rather than measured afterward with a probe — the opening move in the exact same furnace-recipe logic that the alloy-junction, diffused-junction, and planar processes all later refined, each by finding a more correctable, more batchable place to apply it.