How Should HR and Engineering Write Screening Criteria for a VLSI Role?

Engineering writes the evidence and HR writes the process. The engineering lead names three to five things only the person who did the work could explain, and what weak, middling and strong answers sound like. HR turns those into written criteria, applies them the same way to every applicant, and sends an engineer only the top few.

That is the short answer for a head of HR or talent acquisition at a chip design company, with the engineering head in the room. The rest of this page covers why a copied job description fails as criteria, who owns which half, a five-step method, the four places criteria usually go wrong, and what AgentR does with criteria once they exist. AgentR is one such tool, so read its section with that in mind.

Why can't HR copy the job description into the screening criteria?

Because a VLSI job description lists tools, flows and nodes, and nearly every applicant lists the same ones. Criteria copied from it test vocabulary, which separates nobody. Criteria have to describe what an owner of the work can explain and a supporter of the work cannot.

The pool is crowded with the same words. Careernet's analysis of job postings at 79 semiconductor design GCCs found that core VLSI skills made up 48% of hiring demand in the first quarter of FY26, with digital design at 15% and verification at 10% (PTI via The Week, September 2025). The same report found overall hiring volume down 22% on the year before. Careernet's chief business officer, Neelabh Shukla, said at the time that "the demand for niche, innovation-led skill sets is accelerating, reflecting a decisive shift from quantity to quality."

Demand has since climbed. A later Careernet report counted open roles at semiconductor design GCCs rising from 2,426 in January 2026 to 3,549 in March 2026 (IANS via Lokmat Times, May 2026). Those are multinational capability centres rather than the domestic mid-size firms this page is written for, but they hire from the same pool of engineers. When many firms chase few people, the applicants who reach you describe their work in the same terms, and a keyword screen cannot tell them apart. Our Paper Tiger Index describes that failure from the candidate side.

What does engineering write, and what does HR write?

Engineering writes what good looks like: the probes and the score bands. HR writes how it is applied: the order, the consistency and the record. Neither can do the other's half well. HR cannot judge a timing answer, and an engineer reading 120 CVs one by one will hold a different bar on the hundredth than on the first.

Engineering owns: three to five probes, a weak, middling and strong description for each, and a decision on which probes are must-haves.

HR owns: the screening order, applying the same criteria to every applicant, recording what was found for each one, and candidate consent and communication.

Both sign off: the final criteria before the role opens. Criteria written after you have read the applicants are easy to fit to the applicant you already like, so lock them first.

How do you turn a VLSI skill into a screening probe?

Replace each skill with a question only the owner could answer, then describe three answer levels. "Experience with physical design sign-off" becomes "which checks failed at sign-off on your block, and what did you change?" A weak answer names the tool. A strong one names the failure, the fix and what the fix cost.

The method has five steps.

1. Pick three to five outcomes the hire must deliver in the first year, such as taking a block from RTL to a clean sign-off.
2. For each outcome, write the owner-only question.
3. Write a weak, middling and strong answer in one sentence each, drawing on your own past hires and near misses.
4. Mark which probes are must-haves and which a good engineer can learn on the job.
5. Decide in advance where each piece of evidence will come from: the application, a structured first conversation, or the engineer's own round.

Here is one worked example for a verification role. It is an illustration for you to replace with your engineer's own wording.

Claim on the CV: built UVM testbenches.

Probe: which part of the testbench did you write, and what bug did it catch that the earlier one missed?

Weak answer: names the methodology and nothing else.

Strong answer: names the component, the class of bug, and what changed in the design afterwards.

Where do VLSI screening criteria usually go wrong?

They go wrong in four ways: too many must-haves, tools standing in for skill, criteria written after the CVs were read, and criteria that ignore work under NDA. The fourth is specific to chip hiring and easy to miss.

Too many must-haves

A list of ten must-haves describes a person who may not exist, and the search stalls for months. Keep the must-haves to the few things the hire cannot do the job without, and treat the rest as preferences you score but do not screen on.

Tools standing in for skill

"Five years of a named tool flow" measures exposure, not ownership. Ask what the candidate decided with the tool, not how long they sat in front of it.

Criteria written after reading CVs

Once the first strong CV is in your head, the criteria drift toward it. Write them, sign them and date them before the role opens.

Work under NDA

A candidate who owned a tape-out usually cannot share the netlist, and often cannot share the block name. Criteria that demand artefacts will screen out some of the best people. Ask for the shape of the work instead: the process node generation, the rough scale, the candidate's role, the problem and the decision. Expect described evidence rather than documents, and plan a referee conversation for the finalists.

Does writing criteria first actually improve the screen?

Probably, with a limit. Sackett and colleagues' 2022 meta-analysis found structured interviews had the highest mean validity of the selection methods it compared (r = .42) (SIOP summary). A structured interview is one where the questions and scoring are fixed before anyone is interviewed, which is exactly what the five steps produce.

The limit is that this is an average across many jobs and many interviewers. It is not a study of chip hiring, and no figure here says what a written rubric is worth for a physical design search. Treat it as a reason to try the method on one role and compare the result with how you hired last time.

What does AgentR do with criteria, and what does it leave alone?

AgentR reads every application in full against the criteria you set, checks claims for consistency, runs a structured first interview scored against your rubric, and gives the team a ranked shortlist with the evidence attached. A person makes every decision, and nobody is auto-rejected.

Concretely, AgentR can draft criteria from the job description, and your engineering lead should edit them, because the draft will not know what separates your best hires from your near misses. Its verification rates each claim as proven, described, mentioned or not found. It is a consistency check. It does not contact employers or institutions and is not a background verification, so see what to check before sending a candidate to background verification in India for that step, and what verification does and does not check for the product.

The limits matter as much. AgentR is in private beta, has no customers to point to and has published no accuracy figures for VLSI or any other specialist field. It cannot judge timing-closure depth or design judgment, so every shortlisted candidate still needs a senior engineer. For NDA-bound work expect "described" more often than "proven". And if your engineering lead already reads every CV and you open two roles a year, a written rubric and a spreadsheet may be all you need.

Questions people ask

How do you evaluate RTL, verification or physical design experience from a resume before using an engineer's time?

Turn each claim into an owner-only probe and check whether the application answers it. A candidate who owned the work can usually name the failure and the fix. One who supported the work usually names the tool.

How do you verify tape-out and project claims for semiconductor candidates?

Public sources rarely help, because the work sits behind an NDA. Check that the dates, roles and process node are consistent across the application, ask for the shape of the work, and confirm formal details with a referee or a verifier for finalists.

What are the risks of keyword screening in semiconductor hiring?

It rewards the vocabulary every applicant shares: tool names, flow names, node names. A resume can contain every keyword in the job description and still describe support work, and candidates whose best work is under NDA can be passed over.

Where to go next

If you already have criteria and want to know what a screen can and cannot do with them, read Can AI screen VLSI and physical design candidates?, which also covers cost in rupees as of October 2026.

Write the criteria for one open VLSI role with your engineering lead, then run that role through AgentR. The first role is free, up to 50 applicants and 5 interviews.

See how screening decides who makes the shortlist on the smarter screening page.