channel length modulation

Channel length modulation is the reason a transistor in saturation is not actually a perfect current source, no matter how the textbook equation first presents it. Raise the drain voltage on a MOSFET that is already saturated and the drain current still creeps upward — not because the device entered a new operating region, but because the region of the channel actually doing the work of conduction is physically shrinking underneath the fixed gate.

The physical picture starts from where saturation itself comes from. In a MOSFET's triode region the inverted channel runs unbroken from source to drain; raising drain voltage past the saturation point (V_DS = V_GS − V_TH) pinches the channel off right at the drain edge, and in the simplest model the device then behaves as an ideal current source, independent of V_DS entirely. Real devices do not stop there: pushing V_DS further above that point widens the depletion region near the drain, and the point where the channel actually pinches off retreats from the drain back toward the source. The electrically active channel length shrinks by some ΔL, and resistance falls roughly in proportion — so drain current rises, gently, with drain voltage, instead of staying flat.

The standard long-channel equation patches this in as a single linear correction term, and the parameter λ is the whole story. The textbook saturation current equation multiplies the ideal, V_DS-independent result by (1 + λV_DS), where λ is an empirical channel-length-modulation parameter extracted per process node and per device length. λ is not a universal constant — it scales roughly as 1/L, because ΔL itself depends mostly on voltage and doping, not on the drawn channel length, so the same absolute ΔL is a much larger fractional change in a short device than a long one. This is precisely why channel-length modulation is a bigger problem, not a smaller one, as process nodes keep shrinking L.

Output resistance is where channel-length modulation stops being a second-order correction and becomes a first-order design limit. The finite slope of the saturation I-V curve defines a small-signal output resistance, r_o = 1/(λI_D) — not infinite, as the ideal current-source picture implies, but a real, finite number that directly caps the voltage gain available from a single transistor. An amplifier's intrinsic gain is proportional to g_m·r_o; since r_o shrinks as λ grows and as bias current grows, the same channel-length modulation that is a rounding error for a digital logic gate is often the single dominant limit on how much gain an analog designer can extract from one device.

This is why analog and digital design treat the same physical effect with opposite levels of concern. A digital gate cares about on-current and switching speed, where a few percent of current drift with V_DS barely matters against the much larger swing between on and off. An analog gain stage built from the shortest, fastest transistors a process offers runs directly into the 1/L scaling of λ — which is exactly why high-gain analog building blocks (current mirrors, cascode stages, operational transconductance amplifiers) deliberately use longer-than-minimum channel lengths, trading some speed and area for the r_o a short, fast digital-style device simply cannot provide.

Cascoding is the standard circuit-level answer to a physical effect that cannot be designed out of the transistor itself. Stacking a second transistor on top of the first shields the bottom device's drain from large V_DS swings, holding it near its own saturation edge regardless of what the output node does — which multiplies the effective output resistance by roughly the stacked device's own intrinsic gain. The device-physics limit does not go away; the circuit topology simply stops exposing the sensitive node to the voltage swing that triggers it.

Channel length modulation is the device-physics fine print behind a circuit-level fact every analog designer learns early: gain is not free, and a transistor's own channel does not hold still just because the gate said saturation was reached. The same shrinking that makes logic faster and denser makes this correction larger, which is why the most basic MOSFET equation in any textbook carries a term for it at all.