How Akrida works

Akrida is an AI verification engineer. Here's what it does with your design.

1. Step 1

   You give it the spec and the design.

   The spec says what the block must do. Akrida reads the spec and the port list, never the design's code, so it can't copy the design's mistakes.

2. Step 2

   It plans the tests.

   Each rule in the spec becomes a requirement, with the scenarios needed to prove it.

3. Step 3

   It writes and runs the tests.

   Akrida writes a complete UVM testbench and runs it on your design in a simulator.

4. Step 4

   It sorts real bugs from its own mistakes.

   When a check fails, Akrida works out whose fault it is. It fixes its own test mistakes and runs again until every scenario is reached.

5. Step 5

   You get bug reports with proof.

   Each bug says what should happen, what happened, and the log line that shows it. When a fix lands, Akrida checks again and closes the bug.

Words you'll see

Spec

The document that says what the design must do.

Requirement

One rule from the spec, like "a read returns the last value written".

Scenario

A situation the tests must create to check a requirement, like a write that changes only one byte. Engineers call these coverage points.

Testbench

The test code Akrida writes to exercise your design, in SystemVerilog UVM, the industry standard.

Simulation

Running the design and its tests in software, cycle by cycle, long before any chip is made.

Round

One pass of running the tests, looking at every failure, and fixing or reporting it.

Design bug

A place where the design breaks a rule in its spec.

Akrida's test mistake

A failure caused by an error in Akrida's own test code. Akrida fixes these itself, so you never have to look at them.

Commit

A saved version of the design's code. Akrida checks each new one.

Replay

A past run played back step by step. Free and quick.

Sample data

Made-up example data that shows what a run produces. Real runs replace it.