MIT's 6.780 curriculum — lectures, problem sets, and quizzes — taught interactively by AI using the actual source materials. No classroom, no tuition, no waiting list.
Nine stages transform raw silicon into a working chip. Stages 3–7 repeat 20–80 times per device depending on process node — each repetition building one more layer of the final circuit. This course teaches you how engineers control, measure, and optimize every one of those stages at scale.
Each repeat deposits, patterns, etches, and planarizes one more layer of transistors, gates, or interconnects.
The diagram above shows what a fab does. This course teaches you how engineers keep it under control. Every fab runs thousands of wafer lots simultaneously, with hundreds of tools, across a 6-week cycle time. The math that prevents those processes from drifting out of spec is what MIT 6.780 is about — and it's the same math running the most advanced fabs in the world today.
Semiconductor Manufacturing, taught by Prof. Duane Boning. The AI instructor teaches directly from the original lecture notes, problem sets, and quizzes — paired with modern applications from current semiconductor research.
Statistical Process Control, control chart design, design of experiments, response surface methods, sensors and metrology, run-by-run control, and production scheduling. Graduate-level, application-driven.
New fab technicians, engineers transitioning into semiconductor manufacturing, engineering students, and anyone who wants to understand how the world's most complex industrial systems are kept in spec.
Each prompt below launches a full interactive session in Claude. The AI pulls the actual MIT materials from the PUBLICMCP library, teaches the concept, quizzes you, and won't advance until you've got it. Move at your pace.
Click any week to expand. Copy the prompt, paste it into Claude.ai with the library connected, and start the session. Problem sets follow each lecture block. Quizzes fall at midterm and near the end.
This course was built by a single developer to demonstrate what AI-guided technical training looks like when the AI can retrieve real course materials instead of recalling them from memory. The instructor you talk to isn't a fine-tuned model — it's careful prompt architecture on top of a live retrieval pipeline.
Every MIT OpenCourseWare 6.780 document — lecture notes, problem sets, exams, and papers — converted from PDF into clean, AI-readable Markdown with equations and structure preserved.
Each document runs through an AI enrichment step that extracts a summary, keywords, and topic categories — so a request like "the lecture on nested variance" resolves to the exact file.
The course sits in a searchable library database combining full-text and semantic vector search, alongside 180,000+ items from other open collections. Meaning-based queries work, not just keywords.
An open MCP server exposes the library to Claude — and REST, OpenAPI, and agents.json endpoints open the same content to ChatGPT, Gemini, or any agent. During your session, the instructor fetches the actual lecture pages on demand.
Because the instruction is grounded in retrieval, the AI teaches from the actual problem sets, quotes the real equations, and grades your work against real solutions — it can't drift into plausible-sounding invention. And the pipeline isn't specific to MIT content: the same four steps can turn any organization's training materials into an AI-taught course.
No. The PUBLICMCP library server is free to connect. A free Claude.ai account works — no Pro plan required. You only need to add the library endpoint to your Claude settings once, and it's available in every conversation from that point forward.
The prompts instruct Claude to search and fetch the actual MIT 6.780 lecture notes, problem sets, and quizzes from the PUBLICMCP library. Without the connection, Claude can't access those source materials — you'd get general knowledge instead of the real course content. The library is what makes this a real curriculum, not a chatbot approximation of one.
MIT 6.780 was a graduate-level course. The core content — statistical process control, control chart design, DOE, run-by-run control — requires comfort with probability and basic statistics. Engineers transitioning into semiconductor manufacturing and upper-level engineering undergraduates will find it accessible. The AI instructor adjusts its explanations to your level — just tell it where you're starting from.
The original MIT course ran 13 weeks at 2 lectures per week. Self-paced with AI, each lecture session typically takes 30–60 minutes depending on how deep you go. Problem sets can take 1–2 hours each. You can move through the whole course in a few weeks of focused sessions, or take it over months. The AI waits for you.
Yes. The PUBLICMCP library holds the original MIT 6.780 (Semiconductor Manufacturing, Spring 2003) materials — lecture notes from Prof. Duane Boning, all seven problem sets, two quizzes, and student research reports. The AI pairs each topic with modern content from current semiconductor research to keep it relevant to today's fabs, but the foundational material is the real thing.
You can open any week in any order. That said, the curriculum is designed to build — the CUSUM chart in Week 3 uses the probability theory from Week 2, and run-by-run control in Week 9 builds directly on the EWMA concepts from Week 4. If you jump ahead and find yourself lost, go back one or two weeks and fill the gap.
