On Agent-Based Modeling in Semiconductor Supply Chain Planning

Supply chain (SC) planning in the semiconductor industry is challenged by high uncertainties on the demand side as well as a complex manufacturing process with non-deterministic failure modes on the production side. Understanding the complex interdependencies and processes of a supply chain is essential to realize opportunities and mitigate risks. However, this understanding is not easy to achieve due to the complexity of the processes and the non-deterministic human behavior determining supply chain planning performance. Our paper argues for an agent-based approach to understand and improve supply chain planning processes using an industry example. We give an overview of current work and elaborate on the need for integrating human behavior into the models. Overall, we conclude that agent-based simulation is a valuable method to identify favorable and unfavorable conditions for successful planning.

The semiconductor industry is highly competitive with customers from heterogeneous industry sectors and high demand volatility. Rapidly changing environments as well as contracting product life cycles challenge the supply chain planning process (Geng and Jiang 2009). Manufacturing lead-times (months) are much longer than customer order lead-times (Ott et al. 2013). Products and processes increase in intricacy with every cycle. These market properties, amplified by globalization, diversity of variants, and declining manufacturing depth (Beinhocker 2007), result in highly complex supply chain. In addition, competitive pressure in the semiconductor market requires a continuous endeavor for cost reduction. Two common leverage points are technological enhancements and improvements of operational processes. Advancements in the operational processes seem to offer the most promising opportunities for cost reduction (Mönch et al. 2006). At the same time, flexible operational processes are a prerequisite to deal with the volatile characteristics of the semiconductor market and to manage its inherently complex structures.

The resulting uncertainty hampers a comprehensive digitalization of supply chain processes. On a shop floor level the semiconductor industry has mastered these challenges by making use of automation, robotics, statistical process control and other monitoring techniques. Visible examples of automation in today’s fabrication are automated guided vehicles (AGV) and advanced material handling systems (AMHS). On the one hand this increases the importance of protecting such systems from unpredictable events and other uncertainties that cannot be handled by algorithmic automated decision making systems. On the other hand essential innovations have to be induced from outside the physical system, which replaces prior tasks of workers (e.g. supervisors or operators) who are directly in touch with products and the production system. Therefore, the increased agility in the manufacturing process shifted the challenges induced by the volatile and complex semiconductor market and the quest for innovation to the supply chain planning process, which now has to ensure stability and innovation in the uncertain and complex environment. Especially in the realm of supply chain planning processes, human planners remain important to mitigate the influence of uncertainties by their decisions, which cannot be achieved by automated decision making and furthermore, to seek for potential opportunities and innovations.

Current discussions of practitioners and academics (Chien et al. 2016) confirm that human decisions are crucial for planning and control. Although the planning process is supported by IT systems, human judgement is indispensable. The planning process includes different persons in the planning ecosystem and is characterized by complex interaction patterns. Human decision makers often decide both individually and intuitively within groups about capacity utilization, demand adjustments, wafer starts and options to realize opportunities or to mitigate risks. Due to the interactions of individual agents, emergent system properties arise, incorporating uncertainties caused by human behavior (Chien et al. 2016). As we desire and need the human decisions to mitigate risks and realize opportunities they are accompanied with effects such as overreactions as described in the prospect theory (Kahneman and Tversky 1979) or other cognitive biases that may distort successful planning.

Four level hierarchy of supply chain simulation model

Hierarchy of four simulation levels

Agent-based modeling (ABM) supports the analysis of heterogeneous human behavior on planning performance. This method allows for the specification of individual strategies by defining agent types representing individual planning types such as risk-averse behavior vs. risk-taking behavior. In simulation experiments, individual behavior can be varied and systematically analyzed. Therefore, agent-based modeling is an appropriate method to understand the behavior of human decision makers and their interactions. Simulation results may suggest improvements to avoid negative decision strategies and foster positive ones. Investigating supply chain planning with the method of agent-based modeling opens up the opportunity to design interaction processes more streamlined, efficient and less prone to error. This should improve overall planning performance. In this paper we introduce modeling approaches of a semiconductor company, show the experiences gained, and provide an outlook on planned activities.

Related posts