Case study · Turnkey · Taken into mass production

A 16 nm Bitcoin ASIC of our design, from requirements to mass production

We took a SHA-256 ASIC of our own design, as our own product, from requirements and chip design to miners in mass production, turnkey. The hashboards and miners built on it are our design too, made with our EMS partners. It has 200 hashing cores per die, the most published on 16 nm, and its production silicon hashes at a core voltage as low as 0.38 V.

Hashing cores per die, the most published on 16 nm
200
Lowest core voltage, on production silicon; the best seller runs at about 0.40 V
0.38 V
Up to this per die, at 0.48 V
189 GH/s
Chip level with ASICBoost, on first silicon
58.7 J/TH

The program

From the brief to mass production

Our chips on a hashboard of our design, as the miners carried them.
  1. The brief

    A SHA-256 ASIC and the miners around it, for Bitcoin, Bitcoin Cash and Bitcoin SV, hashing at near-threshold voltage.

  2. Architecture

    A deep pipeline that tries one nonce per clock, in an ASICBoost-ready many-core architecture, proven on FPGA before layout.

  3. The full‑custom hash core

    Custom multi-bit flip-flops and adders, foundry libraries re-characterised to ultra-low voltage, and hand placement of the compute core. Its sub-blocks abutted without boundary cells.

  4. First silicon

    On first silicon the chip measured 58.7 J/TH at chip level with ASICBoost, 65.0 J/TH without.

  5. Production silicon

    Two hundred cores in four columns, abutted in pairs: a core voltage as low as 0.38 V, 102.6 GH/s per die there, and up to 189 GH/s per die at 0.48 V.

  6. Test and binning

    Every die graded by test and binning of many-core chips, the subject of our patent.

  7. The hashboard and miner built on the chip

    The hashboards and firmware that bring the chip up, designed in-house, with power supplies developed jointly with our partners.

  8. Mass production

    Our own chips taken into mass production as our own product, and the miners built on them with our EMS partners.

Against the best seller

Beside the best-selling 16 nm Bitcoin ASIC

Its figures are published, or follow from its miner's published rating.

MeasureOursThe best seller
Hashing cores per die200114
Core voltageAs low as 0.38 V, production siliconAbout 0.40 V
Chip-level efficiency without ASICBoost65.0 J/TH, on first siliconAbout 83 to 92 J/TH, derived from its miner's rating

What made the numbers

The methods behind the chip

  1. Custom flip-flops and adders

    Custom multi-bit flip-flops and adders in the hash core, with the compute core placed by hand.

  2. Libraries for near-threshold operation

    For sign-off we re-characterised the foundry's cell libraries for near-threshold voltage.

  3. Patented test and binning

    We hold a patent on test and binning of many-core chips.

  4. A patent on low-noise clocks

    We hold a patent on low-noise clock generation: reducing phase noise in charge-pump PLLs.

  5. Our own bring-up

    Bring-up was done by our team, with outside packaging and analysis partners.

  6. Silicon back into design

    Measurements on the production silicon fed into the follow-on design, refining cell choice and hold margin for dynamic logic at ultra-low voltage.

Questions about the turnkey program

How many hashing cores does the 16 nm ASIC have?

Two hundred per die, in four columns abutted in pairs: the most of any 16 nm Bitcoin ASIC whose full-rate core count is published. The best-selling 16 nm chip has 114.

How low does its core voltage go?

It hashes at a core voltage as low as 0.38 V, at 102.6 GH/s per die, and reaches 189 GH/s per die at 0.48 V, in our measurements on production silicon.

What did the turnkey program include?

Our own design and our own product, carried from the requirements to mass production: the chip's design, first and production silicon, its mass production, and the hashboards and miners built on it, of our design, with our EMS partners.

Your algorithm, our silicon

Tell us the coin, the node and the efficiency you need. We take it from requirements to miners, or design the chip alone.

Our figures are our own chip-level measurements, or the specifications of our designs. Comparisons use figures other makers have published, or that follow from their published miner ratings.