Breker Verification Systems

A 19-person company sitting in the critical path of modern silicon

Somewhere right now, a verification team is staring at a chip with billions of transistors and a deadline that arrived yesterday. The design has to work the first time. Re-spinning a modern system-on-chip costs millions and months. And the test cases that prove it works are, traditionally, written by humans, one painstaking scenario at a time.

Breker Verification Systems exists in that exact moment of pressure. The company builds software that reads what a chip is supposed to do and then generates the multi-threaded, self-checking test cases that hammer on it - automatically, in parallel, across simulation, emulation and actual silicon. It is not a household name. It employs around nineteen people out of San Jose. And it has spent more than two decades sitting in a spot that very few people understand and almost no chip ships without.

Verification was eating chip design alive

Here is the inconvenient truth of the semiconductor industry: designing the chip is the easy part. Proving it works is where projects go to die. By some industry estimates, functional verification consumes the majority of an SoC project's engineering effort. Every new feature multiplies the number of ways the thing can fail. Every processor core you add multiplies the interactions you have to test.

The standard answer, for years, was to throw people at it. Write more testbenches. Hand-code more UVM sequences. Hire more verification engineers than design engineers. It worked, in the way that bailing water out of a boat works - right up until the chips got too complex and the water came in faster than anyone could scoop.

Adnan Hamid had seen this from the inside. Before Breker, he managed AMD's System Logic Division and led the team that built the first test-case generator delivering 100% coverage for an x86-class microprocessor. He had also spent years at Cadence as a system-level verification expert, building solutions for the likes of Texas Instruments, Motorola and General Motors. He knew the bailing-water problem was structural. The fix could not be more people. The fix had to be a machine.

Generate the tests. Don't write them.

In 2003, Adnan Hamid and Maheen Hamid - co-founders, and as it happens, married - made a bet that sounds obvious now and sounded heretical then: that verification intent could be captured once, as a model of what a chip is meant to do, and that test cases could be synthesized from that model the way a logic synthesizer turns RTL into gates. Adnan invented the core technology. Maheen, as COO, ran the operations, finance and the unglamorous machinery that keeps a venture-backed company alive.

They raised a grand total of about $5 million - a rounding error by EDA standards, where the incumbents spend that on a sales kickoff. The last round, a Series A, closed in 2012. What Breker did with that modest sum is the genuinely interesting part.

What the machine actually does

Breker's tools take a description of intended behavior and explore every meaningful path through it - then emit test cases that are multi-threaded, run across multiple processors, and check their own results. The same intent gets reused from block-level simulation all the way to post-silicon bring-up. You write the scenario once. The tool finds the corner cases you would never have thought to type out at 2am.

Two decades, told in dots

The numbers do the arguing

Skepticism is the correct posture toward any vendor claiming to save you time. So here is what Breker reports, and where it comes from. At major semiconductor firms, the company cites a 5X reduction in test composition time and a 4X increase in coverage versus hand-written effort. Its RISC-V SystemVIP is deployed across more than 15 commercial chip projects, and it has engaged with more than 20 RISC-V core producers. Treat the multipliers as the vendor's own figures - but the deployment count is harder to wave away.

The partnerships tell their own story. Breker is a founding member of the Accellera Portable Stimulus Working Group, contributing the technology that became an industry standard. In 2025 it donated a subset of its RISC-V test suite to RISC-V International for future compliance and certification work, with Breker executives chairing related working groups. It has also partnered with Agnisys to generate system-level portable stimulus sequences. Known users include Broadcom and Condor Computing - the latter's president publicly crediting Breker with real advantages over standard verification.

A standard is just a bet that other people will follow

Breker's stated mission is unfussy: "to drive the evolution of verification technology with a single-minded focus on our customers' success." What is striking is how a company this small has shaped tooling this large. Donating core technology to Accellera and RISC-V International is not charity - it is the long game. If the whole industry standardizes on the way you think about verification, you do not need to be the biggest vendor. You need to be the one who got there first and stayed.

The next bet has a name: agentic verification

RISC-V is no longer a research curiosity - it is heading into data centers, cars and AI accelerators, all of which need verification that can stand up to certification. That is squarely Breker's terrain. And the company is already circling the next problem: it has begun publicly asking whether AI can create the missing verification models, and what "agentic verification" - AI agents that test silicon - would actually take. It is the same instinct that started the company in 2003. See the place where humans are doing too much by hand. Build the machine.

Back to that verification team, staring at billions of transistors and a deadline that arrived yesterday. Two decades ago they would have started typing test cases and prayed they thought of everything. Today, increasingly, they describe what the chip should do and let Breker's tools go hunting for the failures - across cores, across coherency, across silicon. The deadline is still yesterday. The boat is still taking on water. But for the first time, something other than a human hand is doing the bailing - and it does not get tired at 2am.

Links, channels & watching

Profile compiled from public sources. Figures such as "5X" and "4X" are company-reported; funding and headcount are approximate per public records.