- Title: [[MIT 2.830J Control of Manufacturing Processes - Lec 1]] - Type: #source/video - Author: - Reference: - Published at: - Reviewed at: [[2021-10-12]] - Links: https://www.youtube.com/watch?v=kC2SEiGaqoA&list=PL7CF97E01FDE7C51A&index=1 --- [[_Media/MIT Control of Manufacturing Processes S08/lecture1.pdf | Lecture 1 Notes]] [web](https://ocw.mit.edu/courses/mechanical-engineering/2-830j-control-of-manufacturing-processes-sma-6303-spring-2008/lecture-videos/lecture1.pdf) ## Texts – Montgomery, D.C., Introduction to Statistical Quality Control, 5th Ed. Wiley, 2005 – May and Spanos, Fundamentals of Semiconductor Manufacturing and Process Control, John Wiley, 2006. ## Project topics - Process diagnosis - Process Improvement - Process Optimization / Robustness ## Main course topics Physical origins of Variation - Process Sensitivities Statistical models (empirical data driven models) - as opposed to physical models Can do effect and causal models and process design via experimentation Process optimization (robustness) - Ideal Operating points ## Manufacturing Process Control Goals of manufacturing processes (optimize / minimize) - Cost (materials, labor, capital, time, setup costs. equipment, maintenance, energy) - Quality - Rate - Flexibility Time is interesting because it spans across process design / control and operational issues (MIT 2.853 manufacturing systems e.g. scheduling, throughput, transport time, interaction of quality and quantity (throughput)). ## Focus Primary focus as techniques are developed - Unit operations (for high-end mfg, every unit has to be operating efficiently, can't just optimize at the end) Secondary - Improving throughput - Improving flexibility - Reducing Cost Monitor, Detect, Compensate ## Typical process control problems - Min feature size on a semiconductor chip has too much variability (e.g. channel size on MOS transistors) => - DNA diagnostic chip (micro-fluidic / polymer device) has uneven flow channels - Operational properties of the chip depend on low variation - (Assembly) Toys that don't fit when assembled at home (fit, geometric fit, variation stack up even though each part is close to nominal) - Reliability of small, dense electrical connectors (variability is relative... how much variability is "good enough") - Airframe skin needs trimming and bending to fit frame - automotive can do better because you can learn from volume - joke: boeing plane has $x$ tons of shims ### Other problems - Variability / durability at changeover => reluctance to changeover, lowered flexibility - 100% inspection rate with high scrap rates or frequent rework => low throughput, high costs - yield is also relative to complexity e.g. 90% yield on microprocessor is good but that would be a terrible yield for something like ballpoint pens or syringes. Shift from defect control to parametric variability control in modern manufacturing. ## Manufacturing Processes - How are they defined - How do they do their thing - How can they be categorized - **Why don't they always get it right?** ![[_Media/MIT Control of Manufacturing Processes S08/Process Components.png]] Changing some properties of a material (work piece) by subjecting it to some process by "equipment" to produce a part. Might be immediate physical force but also the environment the equipment creates around the part e.g. acid etching or anodizing. e.g. Conceptual semiconductor process model Conceptual model => "States and transformations of states" ![[_Media/MIT Control of Manufacturing Processes S08/Conceptual Semiconductor Process Model.png]] > [A General Semiconductor Pro cess Moeling Framework](https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.56.5575&rep=rep1&type=pdf) A simplified version of this model might look something like $wafer \subset environment \subset settings$ ## A process model for control ![[_Media/MIT Control of Manufacturing Processes S08/Process Model for Control.png]] - The equipment acting on the material can be thought of as the process - We can idealize the process as some vector $\mathbf Y$ of observable output features that actually matter where The output feature vector $\mathbf Y$ is a function of the process and it's process parameters $\mathbf Y = \Phi(\alpha)$ - **Question** how do we go about finding the ideal $\alpha$ process parameter values and how do we go about controlling those? - could be as simple as first order partial derivative to quantify sensitivity to change. ### The variation equation ![[_Media/MIT Control of Manufacturing Processes S08/The variation equation.png]] #todo digest notes p 28 onward