Page 32 - EE Times Europe Magazine – November 2023

32 EE|Times EUROPE

GREENER ELECTRONICS | SEMICONDUCTOR MANUFACTURING
Tapping AI for Leaner, Greener Semiconductor

Fab Operations

By Saumitra Jagdale

emiconductor fabs are notorious for their high energy and resource Then we use mixed-integer linear program-
consumption; McKinsey found that power consumption during the ming to further optimize the problem from
that starting point. However, this problem
chip production process can reach 100 MW/hour. Despite the indus- is often too complex for computers to solve
Stry’s relentless pursuit of more efficient chip manufacturing processes within a practical time frame.
“To overcome this complexity, we employ
that would minimize fabs’ environmental impact, a critical factor in that decomposition techniques, which break down
equation—fab scheduling—generally has not been addressed with the required the large optimization problem into smaller
subproblems that can be solved in parallel
rigor. But there’s evidence that tools like generative AI can optimize scheduling and are more manageable in scale,” Potter
to make chip manufacturing more efficient and sustainable. added. “Once we have solutions to these
smaller problems, we combine them to create
Scheduling in a fab can involve mak- has created an AI-powered platform for the final production schedule, which is then
ing decisions at different levels about the optimization. Flexciton’s software combines used in the fabrication process.”
movement of work-in-progress materials. It two conventional approaches to production Flexciton uses a cloud-based hybrid sched-
typically relies on rule-based heuristics and scheduling: heuristics and mathematical uling model whose two key components are
the ability of human managers and tool oper- optimization. a global scheduler and a toolset scheduler.
ators to make dispatch decisions in real time. Heuristics is a rules-based approach The global scheduler focuses on the broader
Typically, however, such an approach focuses that often skips complicated mathematics objective of optimizing the overall manufac-
on making the best decisions for individual to provide quick solutions. Mathematical turing process at the fab level. It considers the
machines, without considering how those optimization is an algorithmic problem- big picture and aims to make decisions that
decisions might affect the entire fab. Some- solving approach to consistently deliver the maximize efficiency and productivity across
times, wise choices made for one machine best possible solution. Heuristics is pragmatic the fab. The toolset scheduler works at a more
can slow down the overall production process but less detailed; mathematical optimization granular level, focusing on the efficient oper-
by causing delays at others, disrupting fab is precise but time-consuming. ation of individual machines or tools within
operations at a cost of perhaps millions of Flexciton’s software tries to harness the the fab. Its main goal is to ensure that wafers
dollars a year. best of both approaches with mixed-integer move through the fab machines smoothly and
linear programming. efficiently. It also strives to follow the priority
USING AI FOR IMPROVED “We begin with a relatively simple rule- ranking that was determined by the global
SEMICONDUCTOR FAB SCHEDULING based algorithm to find a good starting point scheduler.
London-based startup Flexciton is one of the for our scheduling problem,” Flexciton CEO To ensure that even the slightest changes
few organizations that have identified the Jamie Potter told EE Times Europe. “This in a fab element’s operation do not hamper
scheduling challenges in fabs. The company initial solution is what we call a ‘warm start.’ the manufacturing process, Flexciton’s

Flexciton and the semiconductor industry value chain (Source: Flexciton)

NOVEMBER 2023 | www.eetimes.eu