PYTHON · NUMPY

Verify the Machine

Catch a bug in AI-generated RTL by writing a testbench — the verification-first discipline that makes AI-assisted design usable.

01Challenge

An AI copilot generated a 4-bit saturating adder (it should clamp results at 15). It looks right and has a subtle bug. Write a testbench that finds an input where it disagrees with the correct behavior. Never ship unverified RTL.

02Model

The industry is moving fast toward AI that writes hardware — LLMs that emit RTL, tools that optimize the layout. The catch, proven again and again on the RTL-generation benchmarks, is that generated hardware is confidently wrong in ways only testing exposes. So the durable skill is not prompting; it is verification. Given a candidate design, you write a testbench — a set of inputs whose correct outputs you know — and you hunt for a mismatch. The design that survives an exhaustive testbench is trustworthy; the one that doesn't just taught you why the check exists.

03Guided practice

Worked: define the correct reference (min(a+b,15)). Faded: write a testbench that returns input pairs to try. Independent: confirm your tests catch at least one mismatch and name it.

04Feedback

PASS WHEN You caught confidently-wrong generated hardware with a testbench — the discipline that makes AI-assisted design safe to use.

If your run fails, check:

FAIL: your tests missed it — the bug hides where a+b is 16..30; include large inputs like (15,1).
FAIL: reference is wrong — a saturating adder returns min(a+b, 15).

05Retrieve & space

What is unreliable about AI-generated RTL? (It is often functionally wrong in ways only testing reveals.)
What is the durable skill around it? (Verification — writing testbenches that expose mismatches.)
Where did this bug live? (The clamp boundary — s>16 instead of s>15.)

06Mastery & project

Writes a testbench that exposes a defect in generated RTL (L4: verification-first AI-assisted design).

Verified logic is the only kind worth fabricating — the discipline you carry into the tape-out capstone.

← Quantize for the Edge The Cost of Getting Made →

PYTHON · NUMPY

Verify the Machine

Write a testbench that catches the bug in the generated saturating adder.

You read

the arrays and values already in scope

You change

the code you write in each cell

Fixed

the dataset and the checks that grade you