The digital “crypto-currency” Bitcoin surged into the public eye last year, tantalizing enthusiasts with the promise of money unfettered by any government before falling victim to digital heists (see “Crypto-currency Security under Scrutiny”) and short media attention spans.
But $130 million worth of bitcoins are still out there, tended by a dedicated community working to expand their use. More are still being “mined” in a process that rewards digital gold-diggers when their software solves mathematical puzzles involved in verifying transactions and regulating how the currency is used (see “What Bitcoin Is, and Why It Matters”). Now some in that community are taking the expensive step of creating custom silicon chips dedicated to running the software that carries out that process.
“It’s a business opportunity, and also because we believe in Bitcoin and what it can do,” says Josh Zerlan, COO of Butterfly Labs, a Kansas City company that is waiting for its first batch of custom chips to come back from an Asian manufacturer. These chips will be resold to Bitcoin enthusiasts in a line of products that plug into a computer via USB and supercharge its efforts to mine the currency. They range in price from $149 to $29,899, and in early 2013 they will start shipping to over a thousand customers who placed advance orders.
Butterfly has competition from several other companies working on application-specific integrated circuits, or ASICs—the industry term for such specialized chips. The stiffest competition seems likely to come from another U.S. company, New York–based BTCFPGA, and two based in China, Avalon and ASICMiner. All three promise to send out products based on their custom chips early next year.
Having an ASIC fabricated is a highly technical and expensive proposition, typically beginning at the low hundreds of thousands of dollars. That’s primarily because contractors—almost always overseas—charge high prices to operate facilities containing the complex equipment needed to carve chips out of silicon wafers. ASICs are etched from blocks of silicon with microscopic precision. Butterfly Labs says it has paid for chips with features as small as 65 nanometers; BTCFPGA and some others say they have opted for a cheaper 90-nanometer production process.
Butterfly Labs occupies a large facility in Overland Park, Kansas, where it has bought equipment to manufacture products based on its chip, says Zerlan. That includes a pick-and-place machine to solder the 7.5-millimeter-square chips onto circuit boards and three production lines to assemble the final products. Zerlan won’t say how much it all cost, but he says the company paid for the project with profits from previous products and funds from unnamed venture capitalists.
The public voice of ASICMiner, who goes by the name Friedcat online, says that no leading-edge technology is required to speed up the mining process significantly, making the challenge “affordable” for relatively small teams.
“Most people consider ASICs to be a highly capital-intensive industry,” he wrote in a message. “It is not for Bitcoin-mining devices, because even [chip] technology 10 years old is much better than current mining devices.” Engineers who worked on the ASICMiner design have experience creating significantly more complex chips at national labs and startups, says Friedcat.
The emergence of bitcoin-mining ASICs is part of a computational arms race rooted in the complex cryptographic scheme designed by Bitcoin’s anonymous inventor, known only by the online name Satoshi Nakamoto. Under Nakamoto’s rules, which are what allow the currency to function without control by a central bank, miners run free software that communicates over the Internet to maintain a distributed global log of all Bitcoin transactions.
That process is also a competition, with the miner whose software completes the next section of the log—known as a block—being rewarded with newly minted bitcoins. The reward is now 25 bitcoins, currently worth $316. Completing a block requires solving a cryptographic puzzle through brute force, so the faster the software can run, the more likely the miner is to win rewards. Nakamoto’s rules also state that only 21 million bitcoins will ever be released, and at an ever-decreasing rate. That schedule is enforced by making the work of mining bitcoins harder and periodically halving the reward for completing a new block.
As Bitcoin gained popularity, some people began spending thousands on heavily customized “mining rigs” capable of performing mining calculations faster than ordinary PCs. Initially relying on high-powered processors, they soon began making use of GPUs, speedy graphics processors that are also popular with the builders of supercomputers. Small businesses popped up to sell mining rigs and the latest GPUs to other miners.
This spring, that arms race took a more serious turn when Butterfly Labs and some other suppliers moved on to FPGAs, chips with a reconfigurable design. Programming FPGAs to perform mining calculations resulted in significant speed increases. But ASICs, with their fixed design, promise to be at least 100 times faster than FPGAs, and thousands of times faster than GPU-based mining rigs.
“ASICs run faster, consume lower power, and offer better area efficiency over FPGAs,” says Simha Sethumadhavan, an assistant professor of computer science at Columbia University, who challenged students in his hardware class to design their own mining chips. One reason ASICs run so fast is that the same circuits can be fitted into smaller spaces than is possible with an FPGA, he says.
Sethumadhavan notes that the profitability of mining is also determined by Bitcoin exchange rates, electricity costs, and the up-front cost of hardware. However, using ballpark figures makes it possible to illustrate how much more a miner could make using an ASIC. An FPGA-based product that Butterfly Labs sells for $599 could pull in $1.50 worth of Bitcoins a day based on today’s exchange rate, excluding electricity costs. A product with one of its new ASICs inside, priced at $649, is expected to work fast enough to earn $55 a day based on the same criteria.
When ASICs do arrive, they could force many bitcoin miners to give up. Since Nakamoto’s system adjusts the difficulty of mining to keep the rate of production constant, it will become significantly more difficult once the new chips go online. Many small-time miners will find that their outmoded equipment will no longer be able to pay for its own electricity bills. “It will become more of a business,” says Butterfly Labs’ Zerlan. He believes that will help strengthen the Bitcoin economy and enable it to be taken more seriously.
But none of that can happen until Zerlan and his competitors deliver their chips. All the most credible-looking ASIC projects have experienced delays and retreated from initial promises that their products would ship in late 2012, citing technical difficulties.
If the products do make it to customers, the competitors may find themselves caught up in a race to release newer, faster versions. Columbia’s Sethumadhavan says custom ASICs may soon face a challenge from chips for mobile devices with circuits dedicated to performing encryption operations. These chips, expected next year, will probably be designed to a standard higher than the miners can reach and could be used to build powerful mining rigs without ASICs.
Meanwhile, the Bitcoin economy is still based more on speculation than on actual business transactions. Blog host Wordpress became the closest thing to a major brand that accepts bitcoins this month, but although anything from alpaca socks to e-mail hosting can be bought using the currency, most such products are offered by enthusiasts inside the Bitcoin community. Unlike real gold, the bitcoins pursued by those building ASICs may not always glitter.
How a Russian cyberwar in Ukraine could ripple out globally
Soldiers and tanks may care about national borders. Cyber doesn't.
Meet Altos Labs, Silicon Valley’s latest wild bet on living forever
Funders of a deep-pocketed new "rejuvenation" startup are said to include Jeff Bezos and Yuri Milner.
Meta’s new learning algorithm can teach AI to multi-task
The single technique for teaching neural networks multiple skills is a step towards general-purpose AI.
Going bald? Lab-grown hair cells could be on the way
These biotech companies are reprogramming cells to treat baldness, but it’s still early days.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.