Three ways to use a transistor
A single MOSFET has three terminals you care about: gate, drain, and source. Amplifying a signal means putting it in at one terminal and reading the result at another — and the *third* terminal, the one you tie to a fixed voltage, is what names the configuration. Tie the source down and you get a common-source amp. Tie the drain (to the supply) and you get a common-drain, better known as a source follower. Tie the gate and you get a common-gate. Same device, three jobs.
Before any of them amplify, the transistor must be on and in the right region. Analog circuits live in saturation (the active region), where the drain current barely depends on drain voltage but responds strongly to the gate — that responsiveness is exactly what gives you gain. Setting up the steady DC currents and voltages that put the device there is the operating point, and every gain number below is measured *around* that point using the small-signal model: replace the transistor with its tiny linear stand-ins and ask how a wiggle in leads to a wiggle out.
* Bias a MOSFET and read its small-signal parameters M1 d g 0 0 nch W=10u L=0.5u Vgs g 0 0.8 Vds d 0 1.0 .op * solve the DC operating (bias) point .print gm gds * gm = transconductance, ro = 1/gds
Common-source: the gain stage
The common-source stage is the workhorse you reach for when you want voltage gain. The signal goes into the gate; the source sits at a fixed voltage (often ground). A wiggle on the gate of +ΔV pushes the drain current up by gm·ΔV, and that extra current flowing through the load resistance pulls the drain voltage *down*. So a small rise in causes a larger fall out — the stage amplifies and inverts.
How much gain? If the load is the transistor's own output resistance ro (the best a single device can do on its own), the gain is the product of your two handles:
Av (common-source) = -gm * ro * gm depends on how hard you bias it: gm = 2*Id / Vov ( Vov = Vgs - Vth, the overdrive ) * gm * ro is the device's 'intrinsic gain' — the ceiling * for one transistor, often ~20-40 dB in a modern process
The source follower: a buffer
Wire the same transistor the other way — signal into the gate, output taken at the source, drain tied to the supply — and it stops amplifying voltage. The source faithfully *follows* the gate, sitting one roughly-constant Vgs below it. This is the common-drain stage, almost always called a source follower. Its voltage gain is just under 1.
If it does not amplify, why build it? Because it is a buffer. Think of a strong person relaying a whisper: the message is unchanged, but now it can drive a heavy door. A source follower presents a light load to whatever feeds it (high input resistance, a gate draws almost no current) yet offers a low output resistance of about 1/gm, so it can drive a heavy capacitive or low-resistance load that would crush a common-source stage. You park one after a high-gain stage to keep the gain from collapsing when you connect the next block.
Av (source follower) = (gm*ro) / (1 + gm*ro) ~ 1 (a hair under unity) Rout (source follower) ~ 1 / gm (nice and low)
Common-gate
The third arrangement holds the gate fixed and feeds the signal into the source, reading the output at the drain. The common-gate stage is the odd one out: it has a low input resistance (about 1/gm) and is non-inverting. Push current into its source and that same current pops out the drain — it is a current buffer, the mirror image of the source follower's voltage buffer.
On its own a common-gate stage is uncommon, but its behavior is the key to the next idea. Its standout property is what it does looking *up* into the drain: it makes the current there extraordinarily reluctant to change with voltage. Hold that thought — it is exactly the lever the cascode pulls.
The cascode: gain without the cost
Here is the bind. A single common-source stage tops out at gm·ro of gain. The obvious fix — stack a *second* amplifying stage on top — works, but every extra gain stage adds its own noise, burns power, and adds a pole that threatens stability later. You wanted more gain, not a second mouth to feed.
The cascode is the elegant escape. Stack a common-gate transistor on top of your common-source transistor. The lower device still does the amplifying — it still sets gm. But the upper common-gate device acts as a shield: it holds the lower transistor's drain voltage nearly rock-steady, so the lower device's ro stops leaking away gain. Effectively the cascode multiplies the output resistance by roughly the top device's own gm·ro, and gain rides along with it.
single common-source: Av ~ gm1 * ro1
cascode (CG stacked on CS):
Rout ~ (gm2 * ro2) * ro1 ( boosted by ~gm2*ro2 )
Av ~ gm1 * (gm2 * ro2 * ro1)
~ (gm*ro)^2 ( roughly the square! )Gain–bandwidth–swing trade-offs
Every choice above pulls on three things at once: gain, bandwidth, and output swing. You almost never get to maximize all three — push one up and another sags. A seasoned analog designer is really just someone with good instincts for which to spend.
Gain versus bandwidth is the famous one. For a stage driving a load capacitor, the low-frequency gain and the −3 dB bandwidth multiply to a value that stays roughly fixed: the gain–bandwidth product. Crank the gain and the bandwidth shrinks to compensate; widen the bandwidth and the gain drops. You are sliding a bead along a fixed-length string, not stretching the string.
GBW ~ gm / (2*pi*CL) ( set mostly by gm and the load cap ) * trade along a near-constant GBW: * A0 = 100, BW = 1 MHz -> GBW = 100 MHz * A0 = 10, BW = 10 MHz -> GBW = 100 MHz (same string) .ac dec 100 1 1G * sweep frequency to see gain roll off
Output swing is the third corner, and it is where the cascode's bill comes due. Every transistor you stack needs a slice of voltage across it to stay in saturation, so a cascode that doubled your gain also clipped how far the output can move before a device falls out of its happy region. On a low supply this hurts — which is why modern designers weigh the cascode's gain against the swing it costs, sometimes choosing cleverer tricks instead. There is no universally right answer; there is only the right answer *for your spec*.