Device Matching and Thermal Runaway in BJT Pairs

Paralleling BJTs — for higher current handling, current mirrors, differential pairs, or push-pull output stages — looks trivial on a schematic: tie the bases together, tie the collectors together, done. In practice, the exponential I-V law that makes a BJT useful is exactly what makes it dangerous to parallel carelessly. Small, unavoidable differences between "identical" transistors get amplified into large current imbalances, and if those imbalances raise junction temperature in a way that further favors the hotter device, the result is thermal runaway — a positive feedback loop that ends in a dead transistor (or several) within seconds.

Why Matching Fails Even With "Identical" Parts

The collector current in the forward-active region follows the Shockley-derived relation:

Ic = Is·exp(Vbe/Vt)

where Is is the saturation current (device-specific, depends on junction area, doping, and process variation) and Vt = kT/q ≈ 25.85 mV at 300 K.

Two transistors from the same reel, same lot, even adjacent die on a wafer, do not have identical Is. Datasheets typically guarantee hFE within a 2:1 to 3:1 spread across a production lot, and Vbe for a given Ic can easily vary by 10–30 mV between "identical" parts at the same temperature. Because Ic depends exponentially on Vbe, that spread is not a rounding error — it's the dominant effect.

Quantifying the imbalance: if two BJTs share a common base-emitter voltage (as they do when bases and emitters are simply tied together), their currents split according to:

Ic1/Ic2 = exp[(Vbe1 - Vbe2)/Vt] = exp(ΔVbe/Vt)

With Vt ≈ 26 mV, a ΔVbe of just 18 mV gives:

Ic1/Ic2 = exp(18/26) = exp(0.692) ≈ 2.0

So an 18 mV mismatch — well within normal part-to-part spread — produces a 2:1 current split. A 60 mV mismatch (still plausible for unselected parts across temperature) gives exp(60/26) = exp(2.31) ≈ 10:1. One device silently carries an order of magnitude more current than its "partner," dissipating proportionally more power while the other coasts.

The Thermal Runaway Mechanism

The imbalance above is a static snapshot. The real danger is what happens next, because Vbe itself is temperature-dependent:

Vbe(T) ≈ Vbe(T₀) - (2 mV/°C)·(T - T₀)     [at constant Ic]

Vbe has a negative temperature coefficient of roughly −2 mV/°C. Follow the loop for two paralleled BJTs sharing a common Vbe node (bases and emitters tied, no per-device emitter resistors):

1. Device A happens to run slightly hotter (more current, or just sits closer to a heatsink hot spot, or has marginally lower Is).
2. Higher temperature in A lowers its Vbe requirement for the same current — but since Vbe is now fixed by the shared node, A's actual current increases instead (from the exponential law, lower effective threshold at fixed Vbe means more current flows).
3. More current in A means more power dissipation (P = Vce·Ic), which raises A's junction temperature further.
4. Return to step 1, with the imbalance larger each pass.

This is a positive feedback loop with no natural stopping point except device failure (bond wire fusing, junction melting, or secondary breakdown). It is the same underlying mechanism that makes a single BJT's safe operating area shrink at high Vce, but here it's between two devices instead of within one. Meanwhile, the "cooler" device B carries proportionally less current and less power, so it never intervenes — it just watches A die, then inherits the entire load itself.

Key point: thermal runaway between paralleled BJTs is a direct consequence of Vbe's negative temperature coefficient combined with the exponential Ic-Vbe law, and it is not self-limiting unless something in the external circuit actively opposes it.

Emitter Ballasting: The Standard Fix

The standard countermeasure is to insert a small resistor Re in series with each emitter before the emitters are tied together. This converts part of the loop gain from thermal-positive-feedback to a negative-feedback term:

Vbe1 - Ic1·Re = Vbe2 - Ic2·Re   (common base, common emitter-node voltage)

If Ic1 starts to rise above Ic2, the voltage drop Ic1·Re increases, which reduces the effective Vbe driving that transistor's base-emitter junction, pulling Ic1 back down. The resistor converts current imbalance into a self-correcting voltage, rather than letting it feed forward unchecked.

Sizing Re: a practical design rule is to make the voltage drop across Re at the intended operating current comparable to or larger than the expected Vbe mismatch — typically 50–100 mV at nominal Ic, so:

Re ≈ (50 to 100 mV) / Ic(nominal)

Worked example: two power BJTs paralleled to share 4 A total (2 A each nominal), with expected ΔVbe mismatch up to 30 mV.

• Target ballast drop: 100 mV at 2 A → Re = 0.1 V / 2 A = 0.05 Ω.
• Check the self-correction: if device A tries to grab 2.5 A while B drops to 1.5 A (a 1 A imbalance), the differential drop across the two Res is (2.5 - 1.5)·0.05 Ω = 0.05 V = 50 mV — comparable to the 30 mV Vbe mismatch that caused the imbalance, so the loop largely cancels it rather than amplifying it.
• Power cost check: P(Re) = I²·Re = 2²·0.05 = 0.2 W per resistor — small compared to each transistor's Vce·Ic dissipation, an acceptable tradeoff for stability.
• Verification: without Re, the same 30 mV mismatch would have produced exp(30/26) ≈ 3.17:1 current split, i.e., roughly 3.04 A / 0.96 A instead of a near-even 2/2 split — confirming the resistor's stabilizing effect is necessary, not cosmetic.

Practical Design Implications

• Never parallel power BJTs without emitter ballast resistors — this is the single most common failure mode in DIY paralleled output stages and push-pull audio/power designs.
• Matched pairs help but don't eliminate the need for Re — monolithic matched pairs (same die) reduce ΔVbe to a few mV and track temperature together, which relaxes but does not remove the runaway risk, especially if the two halves see different local heatsinking.
• Thermal layout matters as much as electrical layout — mount paralleled devices on a common heatsink with good thermal coupling so a local hot spot in one device doesn't run away independent of its neighbor.
• Derate for worst-case Vbe spread, not typical — datasheet Vbe is a typical value at one test condition; use the min/max spread (or measure a sample from your actual lot) when sizing Re.
• This is also why current mirrors use matched geometry, not two random transistors — in a mirror, Re-style degeneration resistors serve the same matching-and-stabilizing role, trading some output impedance/accuracy for robustness against Vbe mismatch.
• Emitter followers and differential pairs are less exposed — they normally run each device with its own defined emitter current source or degenerate emitter resistor already, which is exactly the fix described here; the vulnerable topology is the naive direct parallel connection for current sharing.

Key Takeaways

• BJT collector current depends exponentially on Vbe (Ic = Is·exp(Vbe/Vt)), so even a 20–60 mV mismatch between "identical" parts produces 2:1 to 10:1 current imbalance when bases/emitters are tied directly.
• Vbe has a negative temperature coefficient (≈ −2 mV/°C at constant Ic), which combined with the exponential law creates a positive feedback loop — thermal runaway — when devices are paralleled without correction.
• Emitter ballast resistors (Re) convert the exponential positive feedback into linear negative feedback by making rising current produce a self-limiting voltage drop; size them so the ballast drop (typically 50–100 mV) dominates over expected Vbe mismatch.
• Matched pairs and shared heatsinking reduce the initial mismatch and help devices track together, but neither replaces the need for ballast resistors in a paralleled power stage.
• Always design for worst-case datasheet Vbe/hFE spread, not typical values, when sizing current-sharing components.