Preparing for a lab defense, a student was asked why raising the collector current makes a BJT more sensitive — why gm should depend on IC at all. The answer was honest: "I don't know." They had gm = IC/VT cold, could plug numbers into it, but couldn't say why the IC was in there. That's an extremely common place to be standing, and it's the whole problem with how the formula is usually taught.
The short answer: because the collector current depends exponentially on the base-emitter voltage, not linearly. gm is the slope of that exponential curve at your operating point. Push the operating point higher (more IC) and you're sitting on a steeper part of the curve, so the same tiny wiggle in VBE produces a bigger wiggle in IC. That "bigger response to the same input" is exactly what transconductance measures. The formula gm = IC/VT is not a separate fact to memorize — it's literally that slope, written out.
If you've only ever seen gm = IC/VT handed to you as a clean result, the IC dependence looks like an accident of the algebra. The definition arrives polished, with the physics already divided out, so there's nothing in it that says why current should set sensitivity. It reads like a coincidence.
It isn't. Here's where it comes from.
The curve is exponential, and gm is its slope
A BJT in forward-active operation follows, to good approximation:
IC = IS · exp(VBE / VT)
where IS is the saturation current (tiny, on the order of 1e-15 A) and VT is the thermal voltage, about 26 mV at room temperature.
Transconductance is defined as how much the collector current changes for a small change in base-emitter voltage — the slope of IC versus VBE at the point where you're biased:
gm = dIC / dVBE
Differentiate the exponential. The derivative of exp(VBE/VT) is (1/VT)·exp(VBE/VT), so:
gm = (1/VT) · IS · exp(VBE / VT) = IC / VT
That's the whole story. gm = IC/VT is what you get when you take the slope of the exponential and notice that IS·exp(VBE/VT) is just IC again. The IC in the formula is not a coincidence — it's a stand-in for "how far up the exponential you currently sit." A bigger IC means you're higher up, where the curve climbs faster, so the slope is steeper. (This is also a reminder that the collector current is set by the supply and bias network, not handed over by the base — IC is where you parked the device, and gm reads off the slope there.)
This is the part the definition hides. On a linear curve, the slope is the same everywhere and the operating point wouldn't matter. On an exponential, the slope grows in proportion to the height — and the height is IC. So sensitivity scales with current.
Numbers you can check
Take VT ≈ 26 mV and bias the transistor at IC = 1 mA.
gm = IC / VT = 1 mA / 26 mV ≈ 38 mA/V
So a 1 mV nudge at the base produces:
ΔIC ≈ gm · ΔVBE = 38 mA/V · 1 mV ≈ 38 µA
A 1 mV change in base-emitter voltage swings the collector current by about 38 microamps.
Now double the Q-point to IC = 2 mA — same transistor, just biased with more current:
gm = 2 mA / 26 mV ≈ 77 mA/V
gm doubled. The same 1 mV nudge now gives:
ΔIC ≈ 77 mA/V · 1 mV ≈ 77 µA
Twice the response from the identical input, because the operating point sits higher on the exponential where the slope is twice as steep. Nothing about the transistor changed except where you parked it on its own curve.
The input resistance rπ moves with it. For a current gain β = 100:
rπ = β / gm
At IC = 1 mA: rπ = 100 / 0.038 ≈ 2.6 kΩ
At IC = 2 mA: rπ = 100 / 0.077 ≈ 1.3 kΩ
Double the current, halve the input resistance. The base draws proportionally more signal current because the device is working harder for the same voltage swing.
Why this changes how you think about the Q-point
The usual reason people give for caring about Q-point placement is headroom: bias somewhere in the middle so the output swing doesn't clip against the rails. That's real, but it's only half the reason.
The other half is gain. A common-emitter stage has voltage gain of roughly -gm·RC. Since gm = IC/VT, the gain is set directly by where you bias:
Av ≈ -gm · RC = -(IC / VT) · RC
Bias at a low IC and gm is small, the slope is shallow, and the stage is quiet even with plenty of headroom. Bias higher and gm climbs, the slope steepens, and the same RC delivers more gain. The Q-point isn't just "don't clip" — it's a gain knob, because it picks the point on the exponential whose slope you're amplifying around.
This is also why gm is so predictable in practice. It doesn't depend on β, on device geometry, or on IS — all of which vary part-to-part and with temperature. It depends only on IC and VT. Set the bias current and you've set the transconductance, full stop. (The flip side is that a Q-point pinned by β instead of by a stable bias network drifts as β changes — which is exactly why good biasing fixes IC, and with it gm.) That stability is a direct consequence of the exponential: the slope is proportional to the height, and you control the height.
The intuition to keep
Picture the IC-VBE curve. It's not a ramp, it's a hockey stick that gets steeper the higher you go. gm is how steep the stick is right where you're standing. Stand low on it (small IC) and a small step sideways in VBE barely moves you up. Stand high (large IC) and the same sideways step throws you up much faster. "More current makes the transistor more sensitive" is just "higher up the hockey stick is steeper." gm = IC/VT is the equation of that steepness.
Check yourself
A BJT is biased at IC = 0.5 mA with VT = 26 mV. You change the bias resistor so the new collector current is IC = 4 mA. By what factor does gm change, and what happens to rπ (with β fixed)?
A) gm stays the same; rπ stays the same — gm is a device constant. B) gm increases by 8×; rπ decreases by 8×. C) gm increases by 8×; rπ increases by 8×. D) gm increases by 3×; rπ decreases by 3× — it scales with the square root.
Correct answer: B.
gm = IC/VT scales linearly with IC. Going from 0.5 mA to 4 mA is a factor of 8, so gm increases 8× (from ≈19 mA/V to ≈154 mA/V). Since rπ = β/gm and β is fixed, rπ drops by the same factor of 8. A treats gm as a fixed device property — the exact misconception this is correcting. C gets gm right but moves rπ the wrong way. D invents a square-root law that isn't there; the relationship is linear because the slope of the exponential is proportional to its height.
Close the gap
"I don't know why gm scales with IC" is the right thing to say out loud, because it points at the one missing piece — the exponential underneath the definition. Once that's in place, gm = IC/VT stops being a rule and becomes a slope you can see. Gradual Learning works by finding the exact spot where a clean formula was handed over with the physics already divided out, and putting it back.