Case study · Design service · GDS out · Entered volume production at the customer
A 7 nm Bitcoin ASIC, designed by us and taken into volume production
Our most advanced mining chip design. We took our full‑custom methods from 16 nm straight to 7 nm and designed a SHA-256 ASIC for Bitcoin with 300 hashing cores per die, 50% more than our 16 nm design, designed for 25 J/TH at chip level at a core voltage of 0.34 V: among the top two 7 nm Bitcoin ASICs on both published figures. We delivered it as GDS, and the customer took it into volume production.
- Hashing cores per die, 50% more than our 16 nm design
- 300
- Designed efficiency at chip level, at 0.34 V
- 25 J/TH
- Designed by us, from the specification to the layout
- GDS out
- Taken into volume production by the customer
- In volume
The design
From 16 nm silicon to 7 nm in volume
Skip the node that does not pay
Our 12 nm study showed that node does not pay off for full‑custom hashing, with custom cells and ultra-low-voltage libraries. So we went from 16 nm straight to 7 nm.
Bring the silicon lessons
Measurements on our 16 nm production silicon refined cell choice and hold margin for dynamic logic at ultra-low voltage.
Custom cells at 7 nm
Standard-cell trade-offs, custom flip-flop design and placement trials, carried out by our engineers from 2019 to 2020.
300 cores per die
A Bitcoin ASIC for the customer's miners, with 300 hashing cores per die, designed as a service from the specification to the layout.
GDS out, volume in
We delivered the design as GDS. The customer built the chips and took them into volume production.
Against the published 7 nm figures
Beside the 7 nm Bitcoin ASICs with published chip-level figures
Ours are the design's figures; the others are their makers' published figures. Chip level, not at the wall.
| Chip | Efficiency at chip level | Core voltage |
|---|---|---|
| Our 7 nm design | 25 J/TH, as designed | 0.34 V |
| The leading 7 nm design of 2020 | As low as 23 J/TH | 0.32 V |
| The leading 7 nm design of 2019 | 30 J/TH | Not published |
What a finer node asks for
Where full‑custom design pays on a finer node
More cores per die
A finer node packs more hashing cores into the die: 300 on our 7 nm design, against 200 on our 16 nm design.
Custom flip-flops
Every stage of a deep hashing pipeline ends in flip-flops, so their design shapes the power. We designed our own at 16 nm and at 7 nm.
Placement by hand
Placement trials at 7 nm, carrying the hand placement of our 16 nm compute core to the finer node.
Patented test and binning
We hold a patent on test and binning of many-core chips.
Questions about this design
What does "GDS out" mean?
We deliver the finished layout, as GDS, to the customer, who orders the masks and wafers from the foundry and runs production. Our 7 nm design went this way, into volume production.
Why go from 16 nm straight to 7 nm?
Our 12 nm study showed that node does not pay off for full‑custom hashing with custom cells and ultra-low-voltage libraries, so we went straight to 7 nm, where our design has 300 hashing cores per die.
How efficient is your 7 nm design?
It was designed for 25 J/TH at chip level, at a core voltage of 0.34 V. The chip-level figures makers have published for 7 nm are 23 J/TH at 0.32 V, 30 J/TH and 55 J/TH, which puts it among the top two. At the wall a miner adds the losses of its power supply and fans.
How many hashing cores does your 7 nm design have?
Three hundred per die, 50% more than the 200 of our 16 nm design.
Related pages
Bring us your next node
Tell us the coin, the node and the efficiency you need. We design the ASIC and deliver GDS for your fab, or take it all the way to miners.
Our figures are our own chip-level measurements, or the specifications of our designs.