Design of Experiments in the real world: Lessons from automotive and semiconductor industry

As product design engineers, we are constantly faced with the challenge of making decisions in the presence of uncertainty. Whether we're launching a new product, tackling reliability issues, improving product performance, or juggling competing design constraints, numerous variables can shape the final outcome. Relying on intuition, trial-and-error testing, or OFAT (one factor at a time) testing can be time-consuming, expensive, and may fail to reveal the true product behavior.Design of Experiments (DOE) is one of the most powerful tools used in product development because it enables engineers to understand the relationship between design variables and product performance. Rather than changing one factor at a time, DOE allows multiple parameters to be evaluated simultaneously. It helps identify critical factors, interactions, and optimize variables with significantly fewer runs. This structured approach reduces development time, minimizes testing costs, and improves confidence in engineering decisions. My experience as a product development engineer using DOE across multiple industries, ranging from oil and gas to automotive to semiconductor, the goal was to answer these practical questions: Why is this component failing? Which design parameter matters the most and the sensitivity of the parameter to response? Are there any design interactions that we do not understand? How much margin do we actually have? Which combination of variables gives the best performance? DOE is a decision-making tool rather than a statistical exercise. Automotive Industry: Reducing Connector Mate Force One of my earliest applications of DOE involved an automotive connector program where I was responsible for lowering the terminal mate force. The objective was clear - reduce the force required to mate the connector while maintaining electrical performance, contact reliability, and mechanical durability defined by USCAR and other automotive specifications. A typical contact terminal is shown in the figure below. Receptacle terminal, Source: Molex Blade terminal, Source: Molex The challenge was that many aspects of the terminal interface design were already fixed. Tooling had been released based on the standard interface defined in the previous program, but the connector failed to meet the low-mate-force requirement. A complete redesign was not an option. Any solution had to work within the boundaries of an existing product tooling. To understand the available opportunities, I first investigated the contact physics at the terminal interface. Mating force was influenced by contact normal force, friction between mating surfaces, lubrication characteristics, material properties, and wear behavior. While some geometric parameters were locked, several variables affecting contact mechanics remained available for optimization. A DOE was developed to evaluate the influence of normal force, lubrication conditions, plating thickness variation, and selected terminal features that could still be modified without requiring new tooling. The goal was not simply to find a lower mating force. The design also needed to maintain sufficient contact force throughout the connector's service life and remain robust against wear and environmental exposure. The study revealed that friction-related effects contributed more to mating force than initially expected. Interactions between lubrication and contact normal force proved particularly important. By understanding these relationships and optimizing the variables that remained under engineering control, the team achieved a 38% reduction in mating force without requiring a major redesign of the connector system. The project reinforced an important lesson that has carried through my career- DOE is often most valuable when engineers are working within constraints. Rather than searching for an ideal solution on a blank sheet of paper, it helps identify the best solution available within the realities of manufacturing, cost, and schedule. Semiconductor Industry: Accelerating Product Development and Process Optimization The same DOE methodology applies throughout the semiconductor industry, where engineers routinely evaluate complex interactions between mechanical, thermal, electrical, and process variables. Compared with many traditional mechanical design applications, semiconductor product development often involves a much larger design space, tighter performance tolerances and significantly higher costs associated with design iterations and testing. As a result, DOE has become an essential tool for both product development and manufacturing process optimization. DOE is widely used to optimize process parameters such as chamber pressure, temperature, gas flow, RF power and process time, as well as hardware design variables including material selection, component geometry, cooling strategies, sealing methods, and assembly conditions. Depending on the application, engineers may analyze responses such as wafer temperature uniformity, stress, displacement, vibration, thermal deformation, critical dimension (CD) control, defect density, reliability, fatigue life or particle generation. Evaluating these variables simultaneously enables engineers to identify the most influential factors and understand interactions that would be difficult or impossible to detect using one-factor-at-a-time testing. For example, consider the development of a new electrostatic chuck used in a plasma etch chamber. One of the design objectives may be to improve wafer temperature uniformity while maintaining sufficient structural stiffness and minimizing thermal stresses. Several design variables—including cooling channel geometry, interface material thickness, heater power - can all influence performance. Instead of changing one parameter at a time, a DOE can be used to systematically evaluate the combined effects of these variables. The study may reveal, for instance, that interface material thickness has the greatest influence on temperature uniformity, while cooling channel geometry becomes significant only at higher heater power levels. Identifying these interactions early allows engineers to focus on the most impactful design changes while avoiding unnecessary iterations. Beyond process optimization, DOE also plays a key role in improving equipment reliability and reducing manufacturing variation. It is commonly used to evaluate the effects of material properties, manufacturing tolerances, assembly methods, and operating conditions on long-term hardware performance. When combined with simulation tools such as finite element analysis (FEA) or computational fluid dynamics (CFD), engineers can evaluate hundreds of virtual design combinations before building physical prototypes. This simulation-driven approach significantly reduces development time, lowers testing costs, and enables data-driven engineering decisions while delivering more robust and reliable semiconductor products. Using DOE with Simulation: A Simple Mounting Bracket Example Many engineers associate DOE with physical testing. However, during the early design phase, efficient studies can be performed using simulation models. Combining DOE with Finite Element Analysis (FEA) allows engineers to evaluate a large design space before building hardware, reducing both development cost and test time. In this example, I have taken a simple mounting bracket to demonstrate how DOE can be applied with simulation. This example is intended for engineers who are just beginning their DOE journey. Assume the objective is to minimize bracket deflection under a fixed load of 10N applied to the top surface as shown in the figure below. Example problem of load applied on bracket Several design variables may influence the response, but we want to understand both sensitivity and cost tradeoff: Mount height Mount width Sheet metal thickness Instead of changing one variable at a time, a DOE can be created in Minitab to systematically vary all factors across a series of design points. For each DOE run, the corresponding geometry is generated in a CAD or FEA tool and analyzed using a structural simulation. The resulting tip displacement becomes the response variable recorded in the DOE table. The workflow is straightforward: Define the factors and their ranges Create a DOE matrix in Minitab Run an FEA simulation for each design point Analyse results in statistical software. Step 1: Define the factors and their range Let’s consider the following factors and levels for our DOE. We can keep the material as Aluminum for each design variation. Table 1: Factors and levels for mounting bracket DOE example Factor Low High Mount Width 100 mm 200 mm Mount Height 100 mm 150 mm Sheet metal Thickness 1 mm 2 mm Step 2: Create DOE matrix You can create a DOE matrix assuming a full factorial design. As you can tell based on 2 levels for each factor for each level there can be 2^3 = 8 runs. Table 2: DOE matrix Mount Width Mount Height Thickness 100 100 1 200 100 1 100 150 1 200 150 1 100 100 2 200 100 2 100 150 2 200 150 2 Step 3: Run an FEA simulation for each design point Once the DOE matrix has been defined, FEA simulations can be executed for each design combination. For this example, ANSYS was used to perform the structural analysis. ANSYS Workbench uses its built-in parametric modeling and optimization capability. Instead of creating a separate CAD model for every design iteration, geometric dimensions can be defined as parameters and automatically varied according to the DOE matrix. ANSYS then generates the required design points, runs the simulations and extracts the response variables for each configuration. ANSYS DOE setupSimulation plot showing deflection of the mounting bracket using ANSYS structural simulation Step 4: Analyse results in statistical software Once the simulations are complete for all the combinations, statistical software like Minitab can be used to determine which variables have the greatest influence on bracket deflection. The first chart to examine when analyzing DOE results is the Pareto chart. This is one of the most valuable tools because it ranks the effects based on their contribution to the response, allowing you to quickly identify the factors that have the greatest impact on performance. As shown in the figure below, factors B, BC, AB, and ABC appear to have little influence on the response and are therefore considered statistically insignificant. These terms can be removed sequentially, starting with the least significant interaction (ABC) and working upward. After eliminating the non-significant factors and refitting the model, the resulting Pareto chart (Figure 6) clearly highlights only the critical factors and interactions that drive the response. Pareto chart without removing non-critical terms Pareto chart after removing non-critical terms Examining the reduced Pareto chart, it is clear that mount height has a negligible effect on mounting bracket deflection and can be considered non-critical within the studied design space. The analysis shows that the primary factors influencing deflection are mount width and bracket thickness, along with the interaction between these two variables. This indicates that changes in width and thickness have the greatest impact on structural stiffness, while variations in mount height contribute little to the overall response. The next set of plots to examine are the interaction plot and the main effects plot, as they provide deeper insight into how the design variables influence the response. The interaction plot reveals a strong interaction between mount width and sheet metal thickness, with non-parallel lines demonstrating that the effect of mount width depends on bracket material thickness. Specifically, thinner sheet metal (1mm) is more sensitive to changes in mount width, resulting in larger variations in deflection - evident in the steep blue slope for 1mm thickness in the interaction plot, whereas thicker sheet metal (2mm) exhibits much lower sensitivity, indicating a more robust design (flatter red slope). However, this robustness comes at a cost premium, as 2mm sheet metal could be more expensive. The main effects plot complements this analysis by quantifying the independent influence of each factor on deflection. Steeper slopes highlight factors with greater impact, while flatter lines indicate parameters with negligible influence. Together, these visualizations guide engineers in balancing structural performance and economic constraints, thinner sheet metal offers cost savings but demands tighter control over mount width to mitigate deflection sensitivity, while thicker sheet metal delivers consistent performance at a higher material cost, enabling data-driven trade-offs between robustness and budget. Interaction plotMain effect plot Another valuable outcome of a DOE study is the development of a regression equation that mathematically relates the input factors to the response. Once the significant factors and interactions have been identified, the regression model can be used to predict performance without requiring additional simulations or physical testing. A key advantage of this approach is that it allows engineers to evaluate factor levels between the discrete values originally defined in the DOE. Rather than being limited to the high and low settings used in the experiment, the regression equation enables interpolation within the design space, providing greater flexibility for optimization and design decision-making. This predictive capability transforms DOE from a screening tool into a practical engineering model that can be used to explore design trade-offs and identify optimal parameter combinations. Regression equation for bracket deflection The same methodology scales directly to real engineering problems. Instead of a mounting bracket, the model may represent a connector terminal, a speaker enclosure, a semiconductor chamber component or a robotic arm. The simulation outputs may include stress, displacement, temperature, vibration or fatigue life. Regardless of the application, the process remains the same: use DOE to explore the design space and simulation to generate the response data.In summary, DOE is a powerful engineering methodology that enables product development teams to make data-driven design decisions by systematically evaluating multiple variables simultaneously. Through real-world examples from the automotive and semiconductor industries, along with a simulation-based mounting bracket case study, this article demonstrates how DOE helps engineers identify critical design factors, understand interactions between variables, reduce development time, testing costs and optimize product performance. Whether applied to physical testing or finite element simulations, DOE provides a practical framework for developing more robust, reliable and cost-effective engineering solutions.