FinFET

A FinFET (fin field-effect transistor) is a non-planar, three-dimensional transistor in which the conducting channel is raised into a thin vertical "fin" of silicon, with the gate draped over two or three sides of that fin. Wrapping the gate around the channel gives it far better electrostatic control than a flat device, suppressing the leakage currents that plagued aggressive transistor shrinks through the 2000s. This is the switching element packed by the billion into every modern Bitcoin mining ASIC — and in mining, where the product is literally joules converted into hashrate, transistor leakage is not an abstraction. It is wasted wattage on your power bill, around the clock.

Why the fin was necessary

In the older planar transistor, the gate sits on top of a flat channel and controls it from one side only. As transistors shrank, the gate's grip weakened: current leaked underneath the channel even when the device was nominally off, and static power consumption ballooned. The FinFET's answer is geometric — stand the channel up as a fin and let the gate grab it from multiple sides. The gate's control scales with how much channel surface it touches, so the wrapped geometry shuts the transistor off far more completely. Intel commercialized the first FinFETs (branded tri-gate) at its 22nm node in 2012, and the structure became the dominant logic transistor across the 16/14nm, 10nm, and 7nm generations at every major foundry.

Why mining cares about fins

SHA-256 ASICs are judged almost entirely on one number: joules per terahash. A mining chip is a sea of identical hashing cores toggling continuously, so its power budget is dominated by two terms — dynamic power, which falls with the square of supply voltage, and static leakage, which the FinFET specifically attacks. Because a FinFET leaks less when idle and switches cleanly at lower voltage, each process-node step delivered a compounding efficiency win. You can read the industry's FinFET era directly in Antminer efficiency figures: the 16nm S9 generation arrived near 98 J/TH, the 7nm S17/S19 era pushed into the 40s and then below 30 J/TH, and 5nm-class silicon in the S19 XP and S21 generations drove efficiency to roughly 21.5 and 17.5 J/TH respectively. Process technology is not the only variable — architecture, voltage tuning, and cooling all matter — but the fin is a large share of why a modern hashboard does more work on less power than five-year-old iron.

Low voltage, low margins

The same physics shapes life on the repair bench and in firmware. FinFET-era hash chips run their logic at remarkably low core voltages, with many chips sharing each supply rail — which is why voltage control on a hashboard is per hash domain, never per chip, and why underclocking pays so well: lowering voltage attacks the dominant, quadratic term in power draw. It also means the margins are thin. A few tens of millivolts of droop across a domain separates stable hashing from a wall of hardware errors, so clean power delivery and good thermal contact matter more with every node shrink.

From 22nm to the GAAFET handoff

Below roughly the 3nm class, even a three-sided grip is no longer enough — the fin gets too thin to control — so the industry is transitioning to the GAAFET, which surrounds stacked channel ribbons on all four sides. The FinFET's decade of dominance is what kept Moore's Law alive through the 2010s, and mining hardware rode that curve as directly as any product on Earth: every generation of fins bought the network more hashes for the same watts. For a buyer weighing used iron against new, this is the context behind the spec sheet: the efficiency gap between generations is mostly transistor geometry, and no amount of tuning will close it — which is exactly why older machines find their second life in heat-reuse roles where the "waste" is the product.