- Oops!Something went wrong.Please try again later.
Semiconductor manufacturing plants, called fabs, are among the most complex factories on the planet. Each silicon wafer must undergo as many as 1,600 steps under carefully controlled conditions to be turned into a computer chip. It’s part of the reason for the current shortage of semiconductors: It isn’t easy to simply speed up production to meet a surge in demand. And building a new fab can take more than two years and an investment of at least $1 billion, and sometimes many multiples of that.
That's one the reason the world is currently facing a shortage of semiconductors for use in everything from cars to laptops.
To help plug the yawning gulf between supply and demand, wringing more efficiency out of each existing fab is vital. But to date, eeking out even small improvements has been difficult. Just ask Tina O’Donnell, the wafer systems manager for hard-drive manufacturer Seagate Technology. The company has used a variety of advanced software to schedule work on its production equipment: algorithms based on hard coded rules, discovered through years of bitter experience, about what works best, as well its own internally developed optimization algorithms, and advanced simulations. But despite all of this, O’Donnell says, Seagate might be able to improve its production efficiency by only a few percent. Its fabs can still experience bottlenecks at different stages of production, she says, especially if one of the chip factory’s sophisticated machines is down for maintenance.
But recently Seagate’s fab in Derry, Northern Ireland, began using a software package created by a London-based company called Flexciton. The software uses a mathematical technique called mixed integer linear programming (MILP for short). The method has an advantage over even most state-of-the-art machine-learning techniques: It can find not just a better factory scheduling solution, but the best one, with mathematical certainty. The problem is, MILP has been too slow to run (that's the case even with powerful computers), to provide the solution in a time frame that is useful for highly dynamic, modular manufacturing lines, like those in most fabs.
Flexciton figured out a way to crack that problem and can now calculate an updated solution for even the most complex fab in less than 15 minutes. “We break the problem down into a series of subproblems that can be solved, and then combine them back up to create the optimal schedule,” says Jamie Potter, Flexciton’s cofounder and chief executive officer.
Using Flexciton at its Derry fab, Seagate was able to achieve a 10% improvement in efficiency in a proof of concept project. It has now rolled the system out to about 60% of the machines in the fab and plans to use Flexciton’s software to schedule its entire production there by the end of the year. “It was amazing,” O’Donnell says of Flexciton’s software. Even if Seagate only sees half the level of improvement it achieved in the proof-of-concept when it rolls Flexciton out to the entire fab, she says it will still represent “a step change in our throughput.”
Potter says that in an average large fab Flexciton will probably increase efficiency—as measured by cycle time, or the amount of time it takes for the factory to produce each semiconductor—7% to 10%, which would translate to $3 million to $5 million in savings per month. Those kind of numbers are driving a flood of interest in Flexciton’s software, Potter says.
Improving production scheduling at fabs was crucial for semiconductor makers, even before the recent chip shortage. That’s because as semiconductors have packed more and more circuitry into ever smaller, thinner silicon wafers, the complexity of creating those chips has skyrocketed. A decade ago it would have taken about 200 different production steps to produce the most common chips; today it often takes more than 1,000. What’s more, the most efficient producers have been able to capture the lion’s share of the market, according to a report from consultants McKinsey & Co.: TSMC, the Taiwanese semiconductor giant that dominates the industry, with an 84% share of global semiconductor revenue, also ranks at the top of the industry by a number of manufacturing efficiency metrics.
Its not just customers that are beating a path to Flexciton’s door. The startup has also caught investors’ attention. Today it announced the completion of a £15 million ($20.6 million) financing round lead by London-based Saras Capital. Also participating in the funding were Backed VC, Chalfen Ventures, Join Capital, Entrepreneur First, Vestey Brothers, SCM, Romulus, and several angel investors.
Potter says the company plans to use the new money to hire up, adding to its current headcount of 40, so that it can serve more semiconductor customers. He also says he sees a role for the company’s software in related areas: for instance, in chip design, where there is an opportunity for thinking about how a semiconductor can be manufactured most efficiently in a particular fab could become an important consideration in configuring the chip.
The software could also play a role in scheduling almost any complex production process in industries ranging from aerospace to chemicals.
Correction, October 21: An earlier version of this story misidentified the location of Seagate's fab. It is in Northern Ireland, not Ireland. The story has also been updated to clarify the nature of the optimization algorithms Seagate used prior to deploying Flexciton.
More tech coverage from Fortune:
Just how massive Amazon has grown during the pandemic, in 8 charts
“Gone too far”: Meet the Dutch chips giant that Silicon Valley loves and Biden fears
Alibaba CEO defends $15.5 billion donation to China’s ‘common prosperity’ drive
Leak reveals how Netflix measures wins and losses. But is it relevant?
4 key products just unveiled at Apple’s MacBook Pro event
This story was originally featured on Fortune.com