JOVANA
Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

Radios on Silicon: Why RFIC Is Different

Forty years ago a Wi-Fi radio would have filled a shoebox with coils, crystals and hand-tuned cans. Today the entire thing — receiver, transmitter, frequency synthesizer and all — hides inside a speck of silicon smaller than a grain of rice, sipping milliwatts so your phone lasts all day. This guide opens the RFIC track by asking the obvious question: why cram a whole radio onto one [[cmos|CMOS]] die, and what makes that job stubbornly harder than any digital chip you've met? We'll lay out the transceiver chain end to end and name every block you'll spend the rest of the track mastering.

From a shoebox of parts to a speck of silicon

Open up a 1980s radio and you find a landscape of discrete parts: a metal-can oscillator you could tune with a screwdriver, fat air-wound coils, ceramic filters the size of breath mints, and a transistor or two soldered onto a board the size of your palm. Every block was a separate component you could pick up with tweezers. It worked — but it was big, expensive to assemble, thirsty for power, and every unit had to be tuned by hand on the production line. You could never have put one in your pocket, let alone billions of them into the world's phones.

An RFIC — a radio-frequency integrated circuit — takes that whole landscape and shrinks it onto a single chip, built from the very same CMOS transistors that make your laptop's processor. Instead of soldering a hundred parts together, a chip designer *draws* them as patterns on silicon, and a fab prints millions of identical copies at once. The economic argument is brutal and decisive: integration crushes cost, size, and power all at the same time. That single fact is why nearly every wireless thing you own — phone, earbuds, smartwatch, the Wi-Fi router on your shelf, the temperature sensor in a smart home — has a radio living on a chip.

The transceiver chain, end to end

A radio that both listens and talks is a [[ic-transceiver|transceiver]] — transmitter plus receiver. The cleanest way to hold the whole subject in your head is to picture it as two pipelines meeting at the antenna, with traffic flowing in opposite directions. Let's walk a signal down each one. On receive, a whisper of radio energy arrives at the antenna and must be coaxed, step by step, into clean digital bits. On transmit, clean digital bits must be built back up into a powerful wave the antenna can fling across the room.

RECEIVE  (signal flows left -> right, getting cleaner & slower)

  ((( ~ )))      weak GHz signal, buried in noise
      |
   [ANTENNA]
      |
    [LNA] ........ amplify GENTLY, add as little noise as possible
      |
   [MIXER] <----- [VCO/PLL]   slide GHz down to a low frequency
      |
   [FILTER] ..... keep our channel, throw away the neighbours
      |
    [ADC] ....... turn the wave into numbers
      |
  0110 1001 ...   bits for the digital baseband


TRANSMIT (signal flows left -> right, getting stronger & faster)

  0110 1001 ...   bits from the digital baseband
      |
    [DAC] ....... turn numbers back into a wave
      |
   [MIXER] <----- [VCO/PLL]   slide low frequency UP to GHz
      |
    [PA] ........ amplify HARD, deliver real power
      |
   [ANTENNA]
      |
  ((( ~ )))      strong GHz signal launched into the air
The two halves of a transceiver. The same building blocks — an amplifier, a mixer, a frequency source — appear on both sides, just tuned for different jobs.

Notice the symmetry. The mixer appears on both sides because it is the frequency elevator of the radio — riding the same physical effect up or down. The VCO/PLL (a voltage-controlled oscillator stabilised by a phase-locked loop) is the chip's tuning dial: it generates the precise reference tone that decides *which* channel you're on. The amplifier at the front of receive is special — it must be quiet — and the amplifier at the end of transmit is special too — it must be strong. Those two opposite personalities get their own names and their own rungs later in this track.

Meet the building blocks

Four blocks do almost all the heavy lifting in a radio. Get a feel for what each one is *for* now, and every later rung will simply deepen a name you already recognise.

  1. [[ic-low-noise-amplifier|LNA]] (low-noise amplifier). The bouncer at the door. The signal from the antenna can be a *millionth* of a volt — so faint it is almost lost in the random electrical hiss every component makes. The LNA's job is to amplify that whisper while adding as little hiss of its own as humanly possible. It is the first block the signal meets, so its noise sets a floor nothing downstream can ever undo.
  2. [[ic-rf-mixer|Mixer]]. The frequency elevator. A 2.4 GHz wave is far too fast for an ADC to digitise directly. The mixer multiplies the incoming signal by the VCO's reference tone, and the maths of multiplying two sine waves spits out their *difference* frequency — sliding a GHz signal down to a few megahertz the rest of the chip can comfortably handle (and, on transmit, sliding it back up).
  3. VCO / PLL. The tuning dial. The voltage-controlled oscillator makes the on-chip reference tone; the phase-locked loop locks that tone to a rock-stable crystal so it lands on exactly the right frequency and stays there. Together they decide which Wi-Fi channel — or which radio station — you are tuned to, with parts-per-million accuracy.
  4. PA (power amplifier). The shouter. To reach a router across the house, the antenna needs real power — often a hundred milliwatts or more, thousands of times stronger than what the mixer produced. The PA is the last and hungriest block in the transmit chain. It is where most of a phone's transmit battery actually goes, and where heat, efficiency and linearity all fight each other.

Why GHz makes everything hard

Here is the heart of why RFIC is its own discipline. A digital chip lives in a world of 1s and 0s where a signal only has to be 'high enough' or 'low enough' — there's enormous margin, and noise barely matters. A baseband analog chip cares about precise voltages but plods along at audio or megahertz speeds. RF lives somewhere far stranger: it must capture tiny voltages buried in noise, at billions of cycles a second, while obeying the physics of waves. Four things conspire to make this the hardest corner of chip design.

  1. Gigahertz speed. At 2.4 GHz a full cycle lasts about 0.4 nanoseconds. Stray capacitances of a few femtofarads and the finite speed of a transistor — things a digital designer cheerfully ignores — now dominate the behaviour. The chip is operating near the very limit of how fast its devices can possibly move.
  2. Tiny signals in noise. The antenna may deliver a signal of a microvolt against a background of thermal hiss. The whole art of the receiver is preserving the signal-to-noise ratio — which is why the LNA's own noise is treated as a sacred, irreversible quantity.
  3. Impedance matching to 50 ohms. At GHz, a wire is no longer just a wire — it is a transmission line, and energy *reflects* off any mismatch like a wave bouncing off a wall. The industry standardised on 50 ohms so every block hands its power cleanly to the next. Get the match wrong and your precious signal echoes back instead of moving forward.
  4. On-chip passives are large and lossy. A digital designer never draws an inductor. An RF designer cannot avoid one — and an on-chip inductor is a sprawling metal spiral that can dwarf a thousand transistors in area, yet stores energy poorly because the silicon underneath steals it. Capacitors and the matching networks they build are similarly bulky and imperfect. Much of RFIC craft is wringing performance out of these reluctant components.

A concrete walk: a 2.4 GHz Wi-Fi receiver

Let's make it real. Your router, three rooms away, transmits a 2.4 GHz Wi-Fi signal. By the time it crawls through two walls and reaches your phone's antenna, it has faded to something on the order of a few microvolts — a thousandth of the voltage your USB port would consider 'off'. Watch how the chain rescues it.

  ~2 uV at 2.400 GHz  (signal-to-noise: just barely positive)
        |
     [ANTENNA] --- matched to 50 ohms so nothing reflects
        |
     [LNA]  gain ~ x60,  adds almost no noise of its own
        |   ->  now ~120 uV, SNR preserved
        |
     [MIXER] <-- VCO/PLL locked to 2.400 GHz exactly
        |   ->  difference frequency: 2.400 GHz - 2.400 GHz
        |        lands the channel near 0 Hz (a few MHz wide)
        |
     [FILTER]  pass our ~20 MHz channel, reject the Bluetooth
        |        and microwave-oven energy crowding the band
        |
     [ADC]  sample the slow waveform into numbers
        |
   01101001 01..   ->  digital baseband recovers your video frame
Every block earns its place: match so nothing reflects, amplify before noise can dominate, mix down so the ADC can cope, filter out the neighbours, then digitise.

Three details make this example sing. First, the LNA goes *first* for a reason: amplify the weak signal before any noisy block downstream gets a chance to drown it. Second, the mixer slides 2.4 GHz down toward zero by beating it against a VCO locked to exactly 2.4 GHz — this 'zero-IF' or direct-conversion trick is what lets a cheap, slow ADC do the digitising. Third, the band is crowded — Bluetooth, other Wi-Fi, even the leakage from a microwave oven all sit at 2.4 GHz — so the filter's selectivity is the difference between a clean video call and a frozen screen.