RF
RF CMOS is what lets a radio transceiver — the part of a phone or radio that actually sends and receives GHz-frequency signals — live on the same silicon as its digital logic, instead of needing a separate specialized chip. Making that work means building high-quality passive components (capacitors, inductors, resistors) directly into a CMOS process that was never originally designed with radio frequencies in mind.
Capacitors are the easiest RF passive to get right, because they don't need to touch the substrate at all. MIM (metal-insulator-metal) capacitors are built as a simple sandwich stacked above the interconnect layers — two metal plates with a thin dielectric between them — giving high capacitance density (pF to nF) and high Q, since the structure never interacts with the lossy silicon underneath. For circuits that need *tunable* capacitance rather than fixed, varactors use a reverse-biased junction whose depletion width — and therefore capacitance — shifts by 3–5x as bias voltage changes; polysilicon/oxide and MOS varactor variants trade off tuning range against Q differently.
Inductors are the hard problem, and the reason is purely geometric. An inductor needs physical loop area to store energy in a magnetic field — there's no way to stack it flat above the interconnect the way a capacitor can be. A spiral trace patterned directly in the metal layers sits right on top of the chip's silicon substrate, and at RF frequencies, the inductor's own changing magnetic field induces eddy currents in that lossy substrate — currents that do nothing useful except dissipate energy as heat. That loss mechanism caps typical on-chip spiral inductor Q at roughly 10–30 at 1GHz, achieving inductances from about 0.5nH to over 10nH depending on trace width, spacing, and turn count. High-resistivity substrates and shielding structures reduce the loss, but can't eliminate the fundamental geometry problem the way MIM capacitors sidestep it entirely.
Tightly coupling two inductors turns them into an on-chip transformer, useful for impedance matching and for building baluns (balanced-to-unbalanced signal converters) — the coupled structure's turns ratio sets its impedance transformation, though its overall Q is bounded by the same substrate-coupling losses affecting each individual inductor.
At GHz frequencies, even ordinary wiring stops behaving like ordinary wiring. Once a signal's wavelength becomes comparable to the length of the trace carrying it, that trace has to be treated as a transmission line rather than an ideal zero-resistance wire — meaning its characteristic impedance (typically targeted at 50Ω) has to be deliberately controlled through trace width and spacing, with differential signal pairs requiring their own controlled differential impedance. Resistors, meanwhile, stay comparatively simple: polysilicon resistors cover most biasing and termination needs at moderate Q, while metal thin-film resistors offer better temperature stability where precision matters more than area.
Put together, these building blocks — high-Q MIM capacitors, substrate-loss-limited spiral inductors, tunable varactors, and impedance-controlled transmission lines — are what let a single CMOS die house both a GHz radio and its digital baseband, instead of requiring a separate specialized RF chip.