A semiconductor company lets new engineers ask questions inside their training material

At this semiconductor company, which works across chip design and manufacturing, a new engineer’s mentor did almost everything: taught them, scored them, found their reading, wrote their plan and answered their questions. Now mentors set the standards and plans, Micius organises the training documents into a reviewed knowledge map, and new hires ask questions inside the material, with every answer pointing back to its source.

At this company, one mentor used to handle nearly every part of a new hire’s training: teaching, scoring, tracking progress, finding material, writing the plan and answering questions. HR started a pilot with graduates and new hires, and split the work from the outset. Mentors keep the role profiles, the scores and the learning plans. Micius reads the training documents, drafts a knowledge map from them and answers questions inside the material, and people review everything it drafts. It doesn’t score anyone or write anyone’s plan.

The training material lived on mentors’ and departments’ own drives: an estimated hundred or more documents in slides, spreadsheets and Word files, plus very long specs and PDFs. Security concerns meant they couldn’t simply be pooled. What a new hire read depended on what their mentor had to hand, and when a cohort finished, the material that worked best stayed in the mentor’s own folders.

Now the training lead imports a department’s files into its own knowledge base, with access set by department. It can also run on site, as the customer required. Micius reads the files one by one and draws out a study outline and key points for each. It splits each document into knowledge points tagged by subject, such as tools, process, excursions or safety; a tool spec several hundred pages long yields hundreds of them. The same method extends to a whole company’s thousands of documents. Under each tag it arranges the points into topics three levels deep, every topic linked to its source documents, and packages the result by role as a knowledge tree: the etch process engineer’s, for example.

The map is a draft until people sign it off. The training lead and mentors check it level by level: are the topics split sensibly, and is each one linked to the right documents? Only then is it published for use in learning plans. If it isn’t right, it goes back to be redone.

Each new hire used to be paired with one mentor, and every mentor had their own yardstick. With no company-wide competency model, two mentors could score new hires in the same role quite differently, and the training lead had no single view of who was falling behind.

Now mentors write a profile for each role, working from the role descriptions in the company’s office system: which skills it covers and how each is weighted. Once reviewed and published, that profile is the yardstick for everyone in the role. Mentors score a baseline, then review at regular intervals, and every score stays in the history. The weak spots show up in the scores, so the mentor picks matching material from the published map, sets due dates and sends the plan to the new hire inside the system.

A new etch process engineer is reading the tool spec and gets stuck in the chapter on endpoint detection. Their mentor is out on the line, so the question used to wait, even though the mentor had answered the same one for the last cohort.

Now the new hire opens the spec from their plan, starts with the outline and key points Micius drew out, and asks right there: when an endpoint triggers early, where should they start looking? Answers draw only on that document and never reach past what the new hire is cleared to see. This one comes back as two points, each with its source; clicking the first opens the passage beside the answer, highlighted, so they can check it for themselves. Where the document is silent, Micius leaves the gap unfilled. When they finish the document, they mark it done and the record is kept.

Mentors now keep profiles, scores and learning plans on one platform and update plans monthly, with fewer files passed around by hand. Because scores, plans and completion records are all logged, the training lead can open any new hire’s history and see where progress is slow. Mentors can then step back from repeat questions and spend their time on the hard ones and on judgement calls. The project tracks how many roles have a published profile, how many new hires have a plan and finish it, how often answers cite the original, and how mentor time and repeat questions change.

Keeping scoring and planning with mentors in this first phase was deliberate, and the records it builds up are what the next pieces will draw on: learning recommendations, exams, and a progress dashboard linked to the company’s office and messaging systems. By the end of a cohort, files that used to sit in mentors’ folders can be part of the knowledge map, ready for the next group of new hires.