Course map
- 5 lessons
Starts with Useful AI Tasks: Start Small and Checkable
- 15 lessons
Starts with Next-Token Prediction and the Sampling Loop
- 13 lessons
Starts with Five Parts of a Prompt
- 14 lessons
Starts with Invented APIs, Flags and Signal Names
- 17 lessons
Starts with Explaining Unfamiliar RTL
- 7 lessons
Starts with Specification to Requirements
- 4 lessons
Starts with APIs: Requests, Responses and Errors
- 5 lessons
Starts with Files, CSV, and JSON
- 6 lessons
Starts with The Retrieval Workflow
- 18 lessons
Starts with Prompts, Scripts, and Agents
- 8 lessons
Starts with Define the Task and Boundaries
- 6 lessons
Starts with What MCP Is
- 6 lessons
Starts with When to Split a Task
- 6 lessons
Starts with Define the Assistant
- 12 lessons
Starts with The Cost Formula
- 18 lessons
Starts with The Nightly Triage Loop
- 10 lessons
Starts with Tokens and Context Questions
- 1 lesson
Starts with AI Skills: Full Course Index
- 1 lesson
Starts with Working with AI Glossary
- 1 lesson
Starts with Working with AI Cheatsheet
About this course
This course is for RTL, design verification (DV), static timing analysis (STA), physical design (PD) and other VLSI engineers starting with AI. You do not need machine-learning mathematics or experience training models. Begin with a small task you can verify, then add tools and automation only after you understand the boundaries. The goal is useful engineering assistance, not an AI research qualification.
What you will be able to do
Write a clear engineering request and supply relevant, permitted evidence.
Separate supported facts, assumptions and proposed checks; reject unsupported sign-off claims.
Use small scripts and exact database queries where deterministic processing is the better choice.
Retrieve specification evidence, preserve versions and recognise unanswerable questions.
Design a bounded read-only agent, understand MCP connections and coordinate a small reviewer team.
Evaluate useful results, human effort, cost and permission boundaries before adopting a workflow.
The fourteen-part learning route
Part | Start here | What you practise |
|---|---|---|
1 | Choose a small task, assess an RTL explanation and keep independent checks. | |
2 | Understand models, context, uncertainty and knowledge limits; deeper tokenisation and attention are optional. | |
3 | State task, evidence, constraints, output and acceptance; improve one request. | |
4 | Check claims independently and use only approved tools and data. | |
5 | Use code, specifications and logs; draft scripts and review their effects. | |
6 | Choose RTL, DV, STA or PD; practise evidence-based analysis without sign-off claims. | |
7 | Understand API boundaries, JSON, diffs and visual evidence before automating. | |
8 | Use files and a small SQL database for exact questions about runs. | |
9 | Retrieve versioned evidence, cite it and recognise when the answer is missing. | |
10 | Distinguish fixed workflows from model-chosen actions and control tool execution. | |
11 | Define safe tools, state, checks, stopping and a guided starter. | |
12 | Connect approved read-only capabilities and package reusable engineering procedures. | |
13 | Divide independent work, merge evidence and compare with a single-agent baseline. | |
14 | Build and assess a regression assistant, then document limitations and handover. |
The route is the recommended order; it does not require completing every older reference lesson. Database internals, attention arithmetic, custom MCP server implementation and operating agents at team scale are extensions. Existing lesson URLs remain available so bookmarks and progress are not lost. The course library currently contains 170 non-hub pages before the glossary, cheatsheet and guide pages; this is a library count, not the number of required lessons.
One engineering story, progressively more capable assistance
Show how the course is organised
The later guided project is a fictional DCPeripheral with an 8-bit configuration register and a depth-four request queue, supported by synthetic specifications, reports and run history. Smaller standalone examples, including the opening counter, teach one idea before integration. They are not automatically the same hardware revision; use each exercise's stated contract.
Choose a starting route
Route | Read and do | Exit artifact |
|---|---|---|
First useful week | Getting started, then the prompt and verify checklists. | A corrected explanation with claim-to-source checks and named unknowns. |
Daily engineering use | Prompting, verification, then one role walkthrough. | A reviewed bug report, test-plan section or report comparison. |
Build a small workflow | Complete the basics, practical foundations, data, RAG and agent design. | A bounded assistant evaluated on known and unseen cases. |
Connect and coordinate | After the single-agent lab: MCP and skills, multi-agent design, then the capstone. | A comparison with a simpler baseline, evidence records and a handover note. |
How to use the exercises
Keep the task, permitted input, answer you assessed, independent check and final decision. Mark a check as planned, hand-derived or executed; a suggested command is not a test result. Authored responses and scripted controllers demonstrate mechanics, not measured AI performance. Use optional live exercises only through approved tools and keep actual failures as well as successes. No example in this course authorises real job submission, RTL changes or external publication.
Optional reference paths
Reference | When it helps |
|---|---|
When a real task is too large, slow or expensive; detailed arithmetic is optional. | |
After the beginner project, when discussing rollout, budgets and incident handling with the team. | |
After doing the exercises; explain decisions from evidence rather than memorising scripts. | |
Glossary and cheatsheet | Look up terms and revisit task, verification and permission checklists. |
Check yourself What does finishing the course establish?
You can scope, build and assess a small assisted engineering workflow with evidence and boundaries. It does not certify RTL, timing closure, CDC safety, privacy compliance or readiness for autonomous production use.