Definition
ASIC (Application-Specific Integrated Circuit) is a custom silicon chip designed to do exactly one job. In Bitcoin mining, that job is computing SHA-256 double-hashes as fast and as efficiently as physics allows — nothing else.
Unlike a CPU or GPU, which trade flexibility for generality, a mining ASIC hard-wires the SHA-256 round function directly into the transistors. There is no instruction decoder, no general-purpose ALU, no operating system on the chip itself — just thousands of hashing cores running the same computation in parallel. That single-mindedness is why ASICs are orders of magnitude faster and more power-efficient than any other hardware for proof-of-work, and why CPU, GPU, and even FPGA mining became economically obsolete years ago.
How an ASIC actually works inside a miner
A modern Bitcoin ASIC is not a single chip you plug in — it is one link in a much larger chain. A typical Antminer-class machine carries three hashboards, and each board holds dozens of ASIC chips wired together. On an S19, that’s 76 chips per board; on an S19 Pro, 114 chips per board. The chips are daisy-chained: a single UART serial line carries job data into the first chip, propagates down the string, and nonce results travel back up the same chain to the control board.
Power delivery is where ASIC mining hardware differs most from a desktop PC. The chips on a hashboard are not each fed an independent voltage. Instead they are grouped into voltage domains wired in series, so the board’s input voltage is divided across the string. An S19 hashboard splits its 76 chips into 38 domains of two chips each; the S19 Pro packs 114 chips into 38 domains of three. This series arrangement is efficient, but it has a hard consequence: voltage is controlled per domain, not per individual chip. One failed chip can pull down or open-circuit an entire domain, which is the root cause behind a large share of dead-board diagnostics.
Chip generations and process nodes
Bitmain’s BM13xx family traces a clear lineage of shrinking process geometry and rising efficiency: BM1387 → BM1397 → BM1398 → BM1362 → BM1366 → BM1368 → BM1370. Each generation packs more hashing cores into a smaller node — the BM1370 used in the S21 Pro family is a 3nm-class part, while earlier BM1366 chips sit at a larger node. Internally a single chip is itself massively parallel: the BM1368, for example, integrates 1,280 cores arranged as 80 large blocks of 16 small cores each. The chip runs a PLL-driven clock that firmware ramps up during init, trading higher frequency for more hashrate at the cost of power and heat.
That trade-off is measured in efficiency (J/TH) — joules of electricity burned per terahash of work. Process-node shrinks are the single biggest lever on efficiency, which is why the newest silicon commands the highest prices and why older chips eventually fall below the break-even line as difficulty climbs. The ASIC computes the proof of work defined by the SHA-256 algorithm; everything else in the machine exists to feed those chips clean power, cool air, and a steady stream of work from the pool.
Why this matters for repair and firmware
- Diagnostics: Because chips share domains in series, a single bad ASIC often kills a whole domain. Knowing the chips-per-domain layout for your model is the first step in board-level repair.
- Firmware: The firmware on the control board is what talks to the ASIC chain — setting clock frequency, reading nonces, and tuning each domain’s voltage. Different chip generations use different command frame formats, so firmware must match the silicon.
- Tuning headroom: Autotuning and undervolting both work at the domain level, calculating safe operating points at runtime rather than applying fixed presets.
Whether you’re running a single open-source board on your desk or a rack of S21s, the ASIC is the irreducible core of the machine — every other component is built to keep that silicon hashing. If you’re shopping for the right chip generation for your power budget and goals, browse current and refurbished hardware on the D-Central miners catalog to compare efficiency across models before you buy.
In Simple Terms
A specialized chip designed exclusively for Bitcoin mining, far faster and more efficient than general-purpose hardware.
ASIC (Application-Specific Integrated Circuit) is a custom silicon chip designed to do exactly one job. In Bitcoin mining, that job is computing SHA-256 double-hashes as fast and as efficiently as physics allows — nothing else.
Unlike a CPU or GPU, which trade flexibility for generality, a mining ASIC hard-wires the SHA-256 round function directly into the transistors. There is no instruction decoder, no general-purpose ALU, no operating system on the chip itself — just thousands of hashing cores running the same computation in parallel. That single-mindedness is why ASICs are orders of magnitude faster and more power-efficient than any other hardware for proof-of-work, and why CPU, GPU, and even FPGA mining became economically obsolete years ago.
How an ASIC actually works inside a miner
A modern Bitcoin ASIC is not a single chip you plug in — it is one link in a much larger chain. A typical Antminer-class machine carries three hashboards, and each board holds dozens of ASIC chips wired together. On an S19, that's 76 chips per board; on an S19 Pro, 114 chips per board. The chips are daisy-chained: a single UART serial line carries job data into the first chip, propagates down the string, and nonce results travel back up the same chain to the control board.
Power delivery is where ASIC mining hardware differs most from a desktop PC. The chips on a hashboard are not each fed an independent voltage. Instead they are grouped into voltage domains wired in series, so the board's input voltage is divided across the string. An S19 hashboard splits its 76 chips into 38 domains of two chips each; the S19 Pro packs 114 chips into 38 domains of three. This series arrangement is efficient, but it has a hard consequence: voltage is controlled per domain, not per individual chip. One failed chip can pull down or open-circuit an entire domain, which is the root cause behind a large share of dead-board diagnostics.
Chip generations and process nodes
Bitmain's BM13xx family traces a clear lineage of shrinking process geometry and rising efficiency: BM1387 → BM1397 → BM1398 → BM1362 → BM1366 → BM1368 → BM1370. Each generation packs more hashing cores into a smaller node — the BM1370 used in the S21 Pro family is a 3nm-class part, while earlier BM1366 chips sit at a larger node. Internally a single chip is itself massively parallel: the BM1368, for example, integrates 1,280 cores arranged as 80 large blocks of 16 small cores each. The chip runs a PLL-driven clock that firmware ramps up during init, trading higher frequency for more hashrate at the cost of power and heat.
That trade-off is measured in efficiency (J/TH) — joules of electricity burned per terahash of work. Process-node shrinks are the single biggest lever on efficiency, which is why the newest silicon commands the highest prices and why older chips eventually fall below the break-even line as difficulty climbs. The ASIC computes the proof of work defined by the SHA-256 algorithm; everything else in the machine exists to feed those chips clean power, cool air, and a steady stream of work from the pool.
Why this matters for repair and firmware
- Diagnostics: Because chips share domains in series, a single bad ASIC often kills a whole domain. Knowing the chips-per-domain layout for your model is the first step in board-level repair.
- Firmware: The firmware on the control board is what talks to the ASIC chain — setting clock frequency, reading nonces, and tuning each domain's voltage. Different chip generations use different command frame formats, so firmware must match the silicon.
- Tuning headroom: Autotuning and undervolting both work at the domain level, calculating safe operating points at runtime rather than applying fixed presets.
Whether you're running a single open-source board on your desk or a rack of S21s, the ASIC is the irreducible core of the machine — every other component is built to keep that silicon hashing. If you're shopping for the right chip generation for your power budget and goals, browse current and refurbished hardware on the D-Central miners catalog to compare efficiency across models before you buy.