B is for Batching
Welcome back to the Flexciton Tech Glossary Series: A Deep Dive into Semiconductor Technology and Innovation. Our second entry of the series is all about Batching. Let's get started!
A source of variability
Let's begin with the basics: what exactly is a batch? In wafer fabrication, a wafer batch is a group of wafers that are processed (or transported) together. Efficiently forming batches is a common challenge in fabs. While both logistics and processing both wrestle with this issue, our article will focus on batching for processing, which can be either simultaneous or sequential.
Simultaneous batching is when wafers are processed at the same time on the same machine. It is very much inherent to the entire industry, as most of the machines are designed for handling lots of 25 wafers. There are also process types – such as thermal processing (e.g. diffusion, oxidation & annealing), certain deposition processes, and wet processes (e.g. cleaning) – that benefit from running multiple lots in parallel. All of these processes get higher uniformity and machine efficiency from simultaneous batching.
On the other hand, sequential batching refers to the practice of grouping lots or wafers for processing in a specific order to minimise setup changes on a machine. This method aims to maximise Overall Equipment Effectiveness (OEE) by reducing the frequency of setup adjustments needed when transitioning between different production runs. Examples in wafer fabrication include implant, photolithography (photo), and etch.
Essentially, the entire process flow in wafer manufacturing has to deal with batching processes. To give a rough idea: a typical complementary metal-oxide semiconductor (CMOS) architecture in the front-end of the line involves batching in up to 70% of its value added steps. In a recent poll launched by FabTime on what the top cycle time contributors are, the community placed batching at number 5[1], behind tool downs, tool utilisation, holds, and one-of-a-kind tools. Batching creates lot departures in bursts, and hence it inherently causes variability in arrivals downstream. Factory Physics states that:
“In a line where releases are independent of completions, variability early in a routing increases cycle time more than equivalent variability later in the routing.” [2]
Successfully controlling this source of variability will inevitably result in smoother running down the line. However, trying to reduce variability in arrival rates downstream can lead to smaller batch sizes or shorter campaign lengths, affecting the effectiveness of the batching machines themselves.
The many complexities of batching
In wafer fabs, and even more so in those with high product mix, batching is particularly complicated. As described in Factory Physics:
"In simultaneous batching, the basic trade-off is between effective capacity utilisation, for which we want large batches, and minimal wait to batch time, for which we want small batches.” [2]
For sequential batching, changing over to a different setup of the machine will cause the new arriving lots to wait until the required setup is available again.
In both cases, we’re talking about a decision to wait or not to wait. The problem can easily be expressed mathematically if we’re dealing with single product manufacturing and a low number of machines to schedule. However, as one can imagine, the higher the product mix, the higher the possible setups and machines. Then the problem complexity increases, and the size of the solution space explodes. That’s not all, there are other factors that might come into play and complicate things even more. Four different examples are:
- Timelinks or queue time constraints: a maximum time in between processing steps
- High-priority lots: those that need to move faster through the line for any reason
- Downstream capacity constraints: machines that should not get starved at any cost
- Pattern matching: when the sequence of batching processes need to match a predefined pattern, such as AABBB
Strategies to deal with batching
Historically, the industry has used policies for batching; common rules of thumb that could essentially be split up into ‘greedy’ or ‘full batch’ policies[3]. Full batch policies require lots to wait until a full batch is available. They tend to favour effective capacity utilisation and cost factors, while they negatively impact cycle time and variability. Greedy policies don’t wait for full batches and favour cycle time. They assume that when utilisation levels are high, there will be enough WIP to make full batches anyway. For sequential batching on machines with setups, common rules include minimum and maximum campaign length, which have their own counterpart configurations for greedy vs full batching.[3]
The batching formation required in sequential or simultaneous batching involves far more complex decisions than that of loading a single lot into a tool, as it necessitates determining which lots can be grouped together. Compatibility between lots must be considered, and practitioners must also optimize the timing for existing lots on the rack to await new arrivals, all with the goal of maximising batch size. [4]
Industrial engineers face the challenge of deciding the best strategy to use for loading batch tools, such as those in the diffusion area. In an article by FabTime [4], [5] the impact of the greedy vs full or near full batch policy is compared. The greedy heuristic reduces queuing time and variability but may not be cost-effective. Full batching is cost-effective but can be problematic when operational parameters change. For instance, if a tool's load decreases (becomes less of a bottleneck), a full batch policy may increase cycle time and overall fab variability. On the other hand, a greedy approach might cause delays for individual lots arriving just after a batch is loaded, especially critical or hot lots with narrow timelink windows. Adapting these rules to changing fab conditions is essential.
In reality, these two policies are extreme settings in a spectrum of possible trade-offs between cost and cycle time (and sometimes quality). To address the limitations of both the greedy and full batch policies, a middle-ground approach exists. It involves establishing minimum batch size rules and waiting for a set duration, X minutes, until a minimum of Y lots are ready for batching. This solution usually lacks robustness because the X and Y values depend on various operational parameters, different recipes, product mix, and WIP level. As this rule-based approach incorporates more parameters, it demands greater manual adjustments when fab/tool settings change, inevitably leading to suboptimal tool performance.
In all of the above solutions, timelink constraints are not taken into consideration. To address this, Sebastian Knopp[6] recently developed an advanced heuristic based on disjunctive graph representation. The model's primary aim was to diminish the problem size while incorporating timelink constraints. The approach successfully tackled real-life industrial cases but of an unknown problem size.
Over the years, the wafer manufacturing industry has come up with various methodologies to help deal with the situation above, but they give no guarantee that the eventual policy is anywhere near optimal and their rules tend to stay as-is without adjusting to new situations. At times, this rigidity has been addressed using simulation software, enabling factories to experiment with various batching policy configurations. However, this approach proved to be resource-intensive and repetitive, with no guarantee of achieving optimal results.
How optimization can help master the batching problem
Optimization is the key to avoiding the inherent rigidity and unresponsiveness of heuristic approaches, helping to effectively address the batching problem. An optimization-based solution takes into account all batching constraints, including timelinks, and determines the ideal balance between batching cost and cycle time, simultaneously optimizing both objectives.
It can decide how long to wait for the next lots, considering the accumulating queuing time of the current lots and the predicted time for new lots to arrive. No predetermined rules are in place; instead, the mathematical formulation encompasses all possible solutions. With a user-defined objective function featuring customised weights, an optimization solver autonomously identifies the optimal trade-off, eliminating the need for manual intervention.
The challenge with traditional optimization-based solutions is the computational time when the size and complexity of the problem increase. In an article by Mason et al.[7], an optimization-based solution is compared to heuristics. While optimization outperforms heuristics in smaller-scale problems, its performance diminishes as problem size increases. Notably, these examples did not account for timelink constraints.
This tells us that the best practice is to try to break down the overall problem into smaller problems and use optimization to maximise the benefit. At Flexciton, advanced decomposition techniques are used to break down the problem to find a good trade-off between reduced optimality from the original problem and dealing with NP-hard complexity.[8]
Many practitioners aspire to attain optimal solutions for large-scale problems through traditional optimization techniques. However, our focus lies in achieving comprehensive solutions that blend heuristics, mathematical optimization, like mixed-integer linear programming (MILP), and data analytics. This innovative hybrid approach can vastly outperform existing scheduling methods reliant on basic heuristics and rule-based approaches.
Going deeper into the solution space
In a batching context, the solution space represents the numerous ways to create batches with given WIP. Even in a small wafer fab with a basic batching toolset, this space is immense, making it impossible for a human to find the best solution in a multi-product environment. Batching policies throughout history have been like different paths for exploring this space, helping us navigate complex batching mathematics. Just as the Hubble space telescope aided space exploration in the 20th century, cloud computing and artificial intelligence now provide unprecedented capabilities for exploring the mathematical world of solution space, revealing possibilities beyond imagination.
With the advent of these cutting-edge technologies, it is now a matter of finding a solution that satisfies the diverse needs of a fab, including cost, lead time, delivery, quality, flexibility, safety, and sustainability. These objectives often conflict, and ultimately, finding the optimal trade-off is a business decision, but the rise of cloud and AI will enable engineers to pinpoint a batching policy that is closest to the desired optimal trade-off point. Mathematical optimization is an example of a technique that historically had hit its computational limitations and, therefore, its practical usefulness in wafer manufacturing. However, mathematicians knew there was a whole world to explore, just like astronomers always knew there were exciting things beyond our galaxy. Now, with mathematicians having their own big telescope, the wafer manufacturers are ready to set their new frontiers.
Authors
Ben Van Damme, Industrial Engineer and Business Consultant, Flexciton
Dennis Xenos, CTO and Cofounder, Flexciton
References
[1] FabTime Newsletter: Issue 24.03
[2] Wallace J. Hopp, Mark L. Spearman, Factory Physics: Third Edition. Waveland Press, 2011
[3] Lars Mönch, John W. Fowler, Scott J. Mason, 2013, Production Planning and Control for Semiconductor Wafer Fabrication Facilities, Modeling, Analysis, and Systems, Volume 52, Operations Research/Computer Science Interfaces Series
[5] FabTime Newsletter: Issue 9.03
[6] Sebastian Knopp, 2016, Complex Job-Shop Scheduling with Batching in Semiconductor Manufacturing, PhD thesis, l’École des Mines de Saint-Étienne
[7] S. J. Mason , J. W. Fowler , W. M. Carlyle & D. C. Montgomery, 2007, Heuristics for minimizing total weighted tardiness in complex job shops, International Journal of Production Research, Vol. 43, No. 10, 15 May 2005, 1943–1963
[8] S. Elaoud, R. Williamson, B. E. Sanli and D. Xenos, Multi-Objective Parallel Batch Scheduling In Wafer Fabs With Job Timelink Constraints, 2021 Winter Simulation Conference (WSC), 2021, pp. 1-11
Explore more articles
View all
Meet Lio, a driving force behind client success as Flexciton's Technical Customer Lead. Discover more about her keen eye for collaboration and passion for innovation in this edition of The Flex Factor.
Meet Lio, a driving force behind client success as Flexciton's Technical Customer Lead. Discover more about her keen eye for collaboration and passion for innovation in this edition of The Flex Factor.
Tell us what you do at Flexciton?
I’m a Technical Customer Lead.
What does a typical day look like for you at Flexciton?
The day is incredibly busy and passes quickly while collaborating with the customer team and other teams at Flexciton, making rapid progress day by day. My focus revolves around ongoing customer work, such as our work at Renesas (analyzing their adherence, checking the Flex Global heat map, and listening to feedback from the client). Additionally, I often work on live demos and PoC projects. The nature of my tasks varies depending on the project stage, ranging from initial data analysis and integration to final stages where I collaborate with sales on deliverables and the story of the final report. While consistently moving forward with projects and meeting weekly targets, we concurrently establish our working methods and standardize processes to improve efficiency for future projects. For lunch, I usually go to Atis, my go-to place for fresh and nutritious meals. People in the office call it a salad, but I consider it the best healthy lunch with the highest ROI.
What do you enjoy most about your role?
I find the most enjoyment in witnessing the impact our product has on customers who need it. It's fulfilling to see their reactions when they share challenges, and I appreciate understanding how Flexciton can collaborate with them, providing that extra element for improvement.
If you could summarize working at Flexciton in 3 words, what would they be?
Creative, Fast, Collaborative.
Given the fast-paced evolution of technology, what strategies do you recommend for continuous learning and skill development in the tech field?
Stay closely connected to the client side. Understanding the technology they're developing and their current tech level (MES and other systems) provides insights into their readiness for Flexciton.
In the world of technology and innovation, what emerging trend or development excites you the most, and how do you see it shaping our industry?
The semiconductor industry's rapid evolution and diversity are fascinating. The competition between TSMC and Samsung Foundry in advanced GAA (gate-all-around) technology is particularly intriguing. While Samsung claims to be ahead, industry voices suggest a bluff with poor yields. The competition is ongoing, and I wonder if TSMC will maintain its lead or if there will be a paradigm shift in the industry.
Tell us about your best memory at Flexciton?
Meeting the Renesas team at their fab in Palm Bay and witnessing one of their operators' reaction to our app was a memorable experience. Kodi, a talented young manufacturing specialist, was genuinely impacted by our technology which was amazing to see in person. After returning home, he even had a piece of code named after him by Amar.
Do you think you have what it takes to work at Flexciton? Visit our careers page to browse our current openings.
AI has unquestionably stood out as the prevailing technological theme of the year. This wave of innovation begs the question: how can the semiconductor industry, which stands at the heart of technological progress, leverage AI to navigate its own intricate challenges?
The dominant technological theme of the year is unmistakably clear: artificial intelligence (AI) is no longer a distant future, but a transformative present. From the startling capabilities of conversational ChatGPT to the seamless navigation of autonomous vehicles, AI is demonstrating an unprecedented ability to manage complexity and enhance decision-making processes. This wave of innovation begs the question: how can the semiconductor industry, which stands at the heart of technological progress, leverage AI to navigate its own intricate challenges?
Complexity-driven Challenges
Semiconductor wafer fabs are marvels of modern engineering, embodying a complexity that rivals any known man-made system. These intricate networks of toolsets and wafer pathways require precision and adaptability far beyond the conventional methods of management. The difficulty of this task is compounded by the current challenges that hinder its dynamic pace: a protracted shortage of skilled labor, technological advancement in product designs, and the ever-present volatility of the supply chain.
The latest generation of products is the pinnacle of complexity, with production processes that involve thousands of steps and incredibly intricate constraints. This complexity is not just a byproduct of design; it is an inherent challenge in scaling up production while keeping costs within reasonable limits.
The semiconductor supply chain is equally complicated and often susceptible to disruptions that are becoming all too common. In this context, the requirement for skilled labor is more pronounced than ever. Running fab operations effectively demands a workforce that's not just technically skilled but also capable of innovative thinking to solve problems of competing objectives, improve processes, and extract more value. No small task in an environment already brimming with complexity.
The Need for AI in Semiconductor Manufacturing
As we delve into Industry 4.0, we find ourselves at a crossroads. The software solutions of today, while advanced, are not the panacea we once hoped for. The status quo has simply reshuffled the problems we face; we've transitioned from relying on shop floor veterans' tacit knowledge and intuition to a dependency on people who oversee and maintain the data in digital systems. These experts manning the screens are armed with MES, reporting, and legacy scheduling software, all purporting to streamline operations. Yet, the core issue remains: these systems still hinge on human intelligence to steer the intricate workings of the fabs.
At the core of these challenges lies a common denominator: the need for smarter, more efficient, and autonomous systems that can keep pace with the industry's rapid evolution. This is precisely where AI enters the frame, poised to address the shortcomings of current Industry 4.0 implementations. AI is not just an upgrade—it's a paradigm shift. It has the capability to assimilate the nuanced knowledge of experienced engineers and operators working in a fab and translate it into sophisticated, data-driven decisions. By integrating AI, we aim to break the cycle of displacement and truly solve the complex problems inherent in wafer fabs management. The potential of AI is vast, ready to ignite a revolution in efficiency and strategy that could reshape the very fabric of manufacturing.
Building AI for the Semiconductor Industry
Flexciton is the first company that built an AI-driven scheduling solution on the back of many years of scientific research and successfully implemented it into the semiconductor production environment. So how did we do it?
Accessing the Data
The foundation lies in data – clean, accessible, and comprehensive data. Much like the skilled engineers who intuitively navigate the fab's labyrinth, AI requires a map – a dataset that captures the myriad variables and unpredictable nature of semiconductor manufacturing.
Despite the availability of necessary data within fabs, it often remains locked in silos or relegated to external data warehouses, making it difficult to access. Yet, partnerships with existing vendors can unlock these valuable data reserves for AI applications.
Finding People Who Can Build AI
The chips that enable AI are designed and produced by the semiconductor industry, but the AI-driven applications are developed by people who are not typically found within the sector. They align with powerhouses like Google and Amazon or deep-tech companies working on future-proof technologies. This reveals a broader trend: the allure of semiconductors has diminished for the emerging STEM talent pool, overshadowed by the glow of places where state-of-the-art tech is being built. Embracing this drift, Flexciton planted its roots in London, a nexus of technological evolution akin to Silicon Valley. This strategic choice has enabled us to assemble a diverse and exceptional team of optimization and software engineers representing 22 nationalities among just 43 members. It's a testament to our commitment to recruiting premier global talent to lead the charge in tech development, aiming to revolutionize semiconductor manufacturing.
AI Needs Cloud
The advent of cloud computing marks a significant milestone in technological evolution, enabling the development and democratization of technology based on artificial intelligence. At the core of AI development lies the need for vast computing power and extensive data storage capabilities. The cloud environment offers the ability to rapidly provision resources at a relatively low cost. With just a few clicks, a new server can be initialized, bypassing the traditional complexities of hardware installation and maintenance typically handled by IT personnel.
Furthermore, the inherent scalability of the cloud means that not only can we meet our current computing needs but we can also seamlessly expand our resources as new technologies emerge. This flexibility provides collaborating fabs with the latest technology while avoiding the pitfalls of significant initial investment in equipment that requires regular maintenance and eventually becomes obsolete.
Security within the cloud is an area where misconceptions abound. As a cloud-first company, we often address queries about data security. It's crucial to understand that being cloud-first does not equate to possessing your data. In fact, your data is securely stored in Microsoft Azure data centers, which are bastions of security. Microsoft's commitment to cyber security is reflected in its employment of more than 3,500 professionals whose job is to ensure that data centers are robust and a fortress for data, offering peace of mind that often surpasses the security capabilities of private data centers.
Effective Deployment of AI in Fabs
The introduction of AI-driven solutions within a fab environment entails a significant change in existing processes and workflows and often results in decision-making that diverges from the traditional. This can unsettle teams and requires a comprehensive change management strategy. Therefore the implementation process must be planned as a multifaceted endeavor and deeply rooted in human collaboration.
A successful deployment begins with assembling the right team—a blend of industrial engineers with intimate knowledge of fab operations, and technology specialists who underpin the AI infrastructure. This collective must not only include fab management and engineers but also those who are the lifeblood of the shop floor—individuals who intimately understand the fab's heartbeat.
When it comes to actual deployment, the process is iterative and data-centric. Setting clear objectives is pivotal. The AI must be attuned to the Fab's goals—be it enhancing throughput or minimizing cycle times. Often, the first output may not align with operational realities—a clear indication of the AI adage that the quality of input data dictates the quality of output. It is at this juncture that the expertise of Fab professionals becomes crucial, scrutinizing and correcting the data, and refining the schedules until they align with practical Fab dynamics. With objectives in place and a live scheduler operational, the system undergoes rigorous in-FAB testing.
Change management is the lynchpin in this transformative phase. The core of successful AI adoption is rooted in the project team's ability to communicate the 'why' and 'how'—to educate, validate, and elucidate the benefits of AI decisions that, while novel, better align with overarching business goals and drive performance metrics forward.
Making AI Understandable and Manageable
The aversion to the enigmatic 'black box' is universal. In the world of fabs, it can be a barrier to trust and adoption —operational teams must feel empowered to both grasp and guide the underlying mechanisms of AI models.
We made a considerable effort to refine our AI scheduler by incorporating a feature that enables the user to influence the objective of what our AI scheduler is tasked to achieve and also to understand the decision. Once a schedule is created, engineers can look through those decisions and inspect and interrogate them to understand why the scheduler made these decisions.
Case Studies: Success Stories of AI Deployment
I firmly believe that we are on the cusp of a transformative era in semiconductor manufacturing, one where AI-driven solutions will yield unprecedented benefits. To illustrate this, let's delve into some practical case studies.
The first involves implementing Flexciton's AI scheduler within the complex diffusion area of a wafer fab—a zone notorious for its intricate processes. We aimed to achieve a trifecta of goals: maximize batch sizes, minimize rework, and significantly reduce reliance on shop floor decision-making. The challenge was magnified by the fab's limited IT and IE resources at the time of deployment. Partnering with an existing vendor whose systems were already integrated and had immediate access to essential data facilitated a rapid and efficient implementation with minimal engagement of the fab's IT team. This deployment led to remarkable improvements: clean tools saw 25% bigger batches, and rework in the diffusion area was slashed by 36%.
Another case study details a full fab deployment, where the existing rules-based scheduling system was replaced with Flexciton's AI scheduler. The goal was to enhance capacity and reduce cycle times. The deployment was staged, beginning with simpler areas starting with metrology tools, through the photolithography area and eventually scaling to the entire fab, yielding a global optimization of work-in-process (WIP) flow. The result was a significant increase in throughput and a staggering 75% reduction in manual flow control transactions, a testament to the AI's ability to autonomously optimize WIP flow and streamline operations.
The Autonomous Future of Semiconductor Manufacturing
In closing, the semiconductor industry stands on the precipice of a new era marked by autonomy. AI technology, with its capacity to make informed decisions without human input, has demonstrated not only the potential for improved KPIs but also a significant reduction in the need for human decision-making. The future of semiconductor manufacturing is one where AI-driven solutions consistently deliver superior production results, alleviating the human workload and steering fabs towards their objectives with unprecedented precision and efficiency.
As we embrace this autonomous future, it becomes clear that the integration of AI in semiconductor manufacturing is not just an enhancement of the status quo but a reinvention of it. With each fab that turns to AI, the industry moves closer to realizing a vision where technology and human ingenuity converge to create a landscape of limitless potential.
Author: Jamie Potter, CEO and Cofounder, Flexciton
Introducing Will, Lead Backend Engineer at Flexciton. Explore his daily tasks, ranging from crafting backend architecture to overseeing the codebase and managing technical debt in this month's edition of The Flex Factor.
Introducing Will, Lead Backend Engineer at Flexciton. Explore his daily tasks, ranging from crafting backend architecture to overseeing the codebase and managing technical debt in this month's edition of The Flex Factor.
Tell us what you do at Flexciton?
I am a lead backend engineer and the software development practice lead. My work involves designing the backend architecture, managing the codebase structure and technical debt, pushing for best practices across the wider engineering team and contributing features to my delivery team.
What does a typical day look like for you at Flexciton?
I usually start my morning by scanning through the production logs from our deployments and seeing if anything looks suspect and in need of an investigation. From there it will depend on what I am focused on for that week which tends to vary a fair amount. The majority of my time is spent coding features or doing large scale design work. Some days I get to spend refactoring and restructuring our codebase, occasionally I will get to work in the devops or optimisation space which I always look forward to. In any given week there will be a handful of ongoing projects at various stages, from architectural designs to software development practice work that needs to be structured and prioritised. No day goes by without me writing at least some code, but there is a fair amount of admin work to do as well.
What do you enjoy most about your role?
The diversity of the work I get to do. My work often overlaps with optimisation and devops so I can find myself speaking the lots of different people throughout the day. There are many opportunities to dive deeper into a topic with various team members willing to support you. Since joining I have worked with terraform, CI pipelines, infrastructure, hardware configuration, optimisation, frontend, customer deployments, database optimisation and management, the application backend and much more.
If you could summarise working at Flexciton in 3 words, what would they be?
Collaborative, Challenging, Diverse.
What emerging technology do you believe will have the biggest impact on our lives in the next decade?
I think the next decade is going to be made great by lots of smaller contributions made across technology from both hardware and software. I don’t have much hope for AGI / useful AGI this decade but there is a lot going on to be excited about. From a hardware perspective we have companies making huge progress in designing chips specifically for model training, and at the other end of the spectrum more companies are putting satellites into orbit to enable global access to high speed internet. AI has fuelled the search in identifying stable structures for proteins and crystals, pushing frontiers of new medicines and treatments, as well as material science. Memory safety in programming languages has started to draw attention from governments too with languages like Rust (and potentially Hylo in the future) likely to lead for memory safe applications. It will be interesting to see how the landscape changes over the next few years and see companies start to shift their codebases over.
What’s the best piece of advice you’d give to someone starting a career in the tech industry today?
I think the best piece of advice would be to throw away any notion of imposter syndrome from the start. Programming, and tech in general, is massive, and its certainly true that the more you know, the more you realise you do not know. Everyone will take a different path throughout their career and find themselves being expert in one topic and (momentarily) hopeless in another. When the topics that you know nothing about come along, its best to embrace that and start finding opportunities to learn. It is important to convince yourself that while you may not be able to learn everything, you could learn anything and find joy in accruing that knowledge as you progress in your career. Bearing this in mind, I would say come into tech because you love it and because you want to learn. There is such as good community across programming languages and industries, anyone who wants to learn can easily find help.
Tell us about your best memory at Flexciton?
I can’t think of one great memory that stands out, but what makes Flexciton great is all the little things that happen week after week such that by Sunday evening, I am looking forward to speaking with my team in Monday standup.