SmartFactory AI Productivity can automatically adjust parameters of dispatch rules in a shorter time

Infineon / Mitsubishi / Fuji / Semikron / Eupec / IXYS

SmartFactory AI Productivity can automatically adjust parameters of dispatch rules in a shorter time

Posted Date: 2024-01-17

By Madhav Kidambi, Applied Materials

Semiconductor front-end factories and semiconductor back-end packaging, testing and packaging factories will deploy local labor dispatch rules and schedules as well as global labor dispatch rules to improve production efficiency. Typically, global rules deploy line balancing algorithms to ensure delivery dates are met and optimize utilization of bottleneck resolution tools. These line balancing algorithms have different parameters that need to be adjusted for a given product mix based on factory status. Today, these parameters are adjusted manually or, in some cases, using simulation modeling capabilities. It is difficult to calculate the impact of these parameters on all equipment, all products and process steps in a factory. Therefore, manually adjusting parameters can have a negative impact on factory KPI indicators, while using simulation techniques to find an optimal set of parameters can be time-consuming. This case details how we use SmartFactory Productivity AI to automatically adjust labor dispatch rule parameters in a significantly shortened time.

See the example in Figure 1 below. Based on global rules, there are four parameters used to determine bottleneck resolution tools and line balancing thresholds to determine whether equipment is under, full, or overloaded based on hours of work in progress (WIP). The table shows the possible value ranges for these parameters. One way to find the optimal values ​​of these parameters for a given plant state is to run a simulation model. In each simulation, we select different combinations of parameter values ​​and measure the resulting KPI indicators, a method known in the literature as "grid search". Each run is a 90-day simulation, and a simulation model takes several days to run a 90-day simulation and measure KPIs such as on-time delivery and production cycle time. This is not practical in day-to-day operations.

Figure 1: Line balancing parameters in global rules

To solve this problem, we deployed SmartFactory AI Productivity and Evolutionary Optimization, combined with simulated annealing methods to simultaneously find the optimal parameters for on-time delivery and production cycle. Figure 2 shows how to deploy the algorithm using SmartFactory AI Productivity (including Simulation AutoSched and Fusion modules as well as RTD and Activity Manager).

Figure 2: Algorithm deployment

Using this approach, we were able to find optimal parameter values ​​in hours instead of days. In each iteration, we change the combination of line balancing parameter values ​​and bottleneck station series. As shown in Figure 3, it previously required 300 iterations of grid search to find the optimal on-time delivery rate of 86.90%, but we only needed 10 iterations of the model run to achieve an on-time delivery rate of 98.83%.

Figure 3: On-time delivery rate modeling

As shown in Figure 4, when iterating with the production cycle as the KPI indicator, the fourth iteration of running the model using our method reached the KPI indicator of 886 hours; in comparison, the KPI indicator of grid search was 992 hours .

Figure 4: Production cycle modeling

As shown in Figure 5, once the optimal setup is obtained, it can be integrated with existing dispatch rules and scheduling applications, and the results consolidated and recalculated in the daily operations of the factory.

Figure 5: Production deployment

Running the local KPI indicator of the total handling volume on the bottleneck equipment, it only took 4 iterations to find the optimal bottleneck threshold and production line balance threshold, reaching a handling volume of 20,181 times.

Simulation optimization is the first step in automating dispatch and scheduling parameters; we plan to further improve this by using reinforcement learning methods to automate these parameters.

Related Reading:

1. Original Chinese text of the blog: SmartFactory AI Productivity can automatically adjust the parameters of labor dispatch rules in a shorter time:

2. The common data model supports rapid deployment of productivity solutions - Dispatching & Reporting (Part 3 of 3):

3. Improve the efficiency of development and deployment of AI and ML solutions:

About Applied Materials SmartFactory® solution

Our comprehensive solutions help semiconductor manufacturers and pharmaceutical manufacturers increase production efficiency, optimize quality, increase output, reduce costs, reduce risks and improve yields. Use Applied Materials SmartFactory and SmartFactory Rx solutions to help you quantify the impact of your KPIs. For more information, please visit

About Applied Materials

Applied Materials, Inc. (Nasdaq: AMAT) is a leader in materials engineering solutions, behind nearly every new chip and advanced display produced worldwide. We help our customers make the possible possible with technologies that can change materials at the atomic level at scale. At Applied Materials, we believe our innovations enable a better future. For more information, please visit

#SmartFactory #Productivity #automatically #adjust #parameters #dispatch #rules #shorter #time