Diffused Base Transistor Process Flow
# Diffused-Base Transistor Device Process Flow: Predeposition Dosing, Drive-In Profile Control & Double-Diffusion Yield
The diffused-base transistor was the first transistor manufacturing process to form both junctions of a device from the same flat wafer surface, using a dopant's own concentration-dependent diffusion front instead of a furnace-timed liquid alloy or a crystal-growth timing trick. The grown-junction process had already shown that a furnace event could set a base width more reproducibly than a hand-placed whisker, and the alloy-junction process had already shown that decoupling junction formation from crystal growth made that event correctable. The diffused-base process, developed at Bell Labs in the mid-1950s, went a step further: predeposit a shallow, saturated layer of one dopant into a wafer surface, then drive it inward with a second high-temperature anneal timed so precisely that a *second*, opposite-type dopant introduced afterward diffuses to a shallower depth and overtakes the first only in a thin, buried layer — the emitter, base, and collector all sit on the same face of the same wafer, their relative depths set entirely by how two successive diffusion fronts outrun or overtake each other.
---
## 1. Why Two Sequential Diffusions, Not One
A grown-junction device commits its entire base width to a single dopant-switch instant inside a crystal pull; an alloy-junction device commits its junction depth to a single furnace soak against a saturating liquid pellet. The diffused-base process instead uses *two* sequential solid-state diffusions performed from the same wafer surface, each with its own independently controllable dose and depth, and relies on a simple ordering fact: whichever dopant was driven in first, for longer, reaches deeper, so a second dopant of the opposite type, introduced afterward and driven for a shorter time, necessarily overtakes the first only near the surface — carving out a thin base region sandwiched between a deep collector-side junction and a shallow emitter-side junction, entirely through *timing order*, not through any mechanical placement at all.
---
## 2. Predeposition and Drive-In Kinetics: Two Diffusion Regimes, One Wafer
Predeposition holds the wafer surface at a fixed, saturated dopant concentration $C_s(T)$ for a short time $t_1$, which is a constant-source boundary condition solved by the complementary error function:
The total dose laid down, $Q = \int_0^\infty C(x,t_1)\,dx$, scales with $\sqrt{D t_1}$ — but because $C_s(T)$ pins the surface value regardless of how long predeposition runs, the predeposition step is really a dose-metering step, not a depth-setting one. The wafer is then sealed (no further dopant source) and driven inward during a second, longer anneal at $t_2 \gg t_1$ — a limited-source condition solved by a Gaussian:
This two-regime structure is the entire point: dose ($Q$, set by predeposition) and depth ($x_j \propto \sqrt{D t_2}$, set by drive-in) become two independently schedulable furnace parameters, instead of being coupled together the way a single-step alloy soak coupled pellet volume to junction depth. A process engineer could now fix the collector-side diffusion's dose and depth first, then run a *second* predeposition-plus-drive-in cycle for the emitter dopant with its own independent $Q$ and $t_2$, confident that a shorter second drive-in would mathematically guarantee a shallower junction — the base width became the *difference* between two independently tunable Gaussian depths, not a single hard-to-split quantity:
Because $W_B$ is now a *subtraction* of two independently measured furnace recipes rather than a direct physical quantity, small, correlated errors in both diffusions' temperature control partially cancel — a furnace that runs slightly hot pushes both junctions slightly deeper together, shrinking the base-width error relative to what either diffusion's absolute depth error would suggest.
---
## 3. Process-Control Data: Base-Width Scatter Across Three Junction-Forming Generations
The historical pattern this graph summarizes is the same one that drove the entire sequence of process replacements covered across these articles: each generation did not necessarily change the underlying device physics, it changed *which variable the manufacturing step was sensitive to*, and chose a variable that was easier to hold constant. The diffused-base process's trick — making the controlled quantity a *difference* between two correlated measurements rather than either measurement alone — is the same common-mode-rejection principle a differential amplifier uses against supply noise, applied here to furnace temperature drift between two sequential anneals on the same wafer.
---
## 4. Predeposition-Drive-In Yield Economics
Because predeposition dose and drive-in depth are recorded, independently schedulable furnace parameters rather than a single irreversible event, the diffused-base process was the first transistor manufacturing flow where an entire *wafer* of devices — not one ingot's worth of dice, and not one furnace boat's worth of alloyed pellets — could be processed simultaneously through planar diffusion, since every device on the wafer sees the identical gas-phase dopant flux and the identical anneal temperature profile at the same time. This is also the first process in the sequence where junction depth could be verified non-destructively mid-flow (four-point-probe sheet resistance after predeposition, before committing to drive-in time), turning yield control from a post-hoc inspection problem into an in-process checkpoint — the direct ancestor of the inline metrology checkpoints that would later become standard across the entire wafer-fab process flow.
---
## Diffused-Base vs. Alloy-Junction vs. Grown-Junction: Where the Two-Step Diffusion Actually Pays Off
| Process Decision | Grown-Junction (1951) | Alloy-Junction (1952) | Diffused-Base (mid-1950s) |
|---|---|---|---|
| Junctions formed from | A single dopant-switch instant inside crystal growth | A single furnace soak against a saturating liquid pellet | Two sequential solid-state diffusions from the wafer surface |
| Depth control variable | Pull-rate timing during growth | Soak time, capped by pellet-volume saturation | Drive-in time, with dose pre-fixed by predeposition |
| Dose and depth coupled? | Fully coupled to pull-rate timing | Coupled through pellet volume and soak time together | Decoupled — predeposition sets dose, drive-in sets depth |
| Base width is... | A direct single-event quantity | A direct single-event quantity | A subtraction of two independently measured depths |
| In-process verification possible? | No — only post-slice inspection | No — only post-alloy inspection | Yes — sheet resistance check after predeposition |
| Batch unit | One ingot per pull | Many dice per furnace boat | Every device on one wafer, simultaneously |
This is why the diffused-base process, not the alloy-junction process that preceded it, became the direct ancestor of the modern planar process: turning junction depth into a *difference* between two correlated, independently schedulable diffusions did not just add a processing step, it converted a single irreversible commitment into a pair of furnace recipes whose errors partially cancel each other — and it was the first process in this entire lineage to treat every device on a wafer as one simultaneously-processed batch rather than one slice of an ingot or one dot on a furnace boat, which is exactly the unit of batching the planar process and everything built on top of it would inherit.