In News
-Sathish Raman
Inside every phone, car, and data center sits hardware that had to survive a punishing gauntlet before it ever shipped: heat, vacuum, vibration, chemical attack, and years of continuous use. Getting that hardware right is now a business measured in the hundreds of billions. Global semiconductor sales reached $791.7 billion in 2025, a record, and every one of those chips was built on physical equipment and components that had to perform flawlessly at a scale where a rounding error becomes a recall. The chips get the headlines. The engineering that makes them manufacturable rarely does.
Sahiti Nallagonda, a hardware reliability expert, reveals how design of experiments is crucial for building robust components in automotive and semiconductor sectors. Discover why this systematic approach, not intuition, prevents costly failures and ensures manufacturability at scale, transforming how serious hardware is built.
Sahiti Nallagonda has spent more than 10 years on that unglamorous side of the problem. A senior mechanical engineer and inventor who designs high-precision hardware for extreme environments, she has moved between two of the most demanding industries there are, automotive and semiconductor, building components that hold their tolerances under conditions that would destroy ordinary parts. In a widely read technical article on how design-of-experiments methodology plays out in the real world, she made the case that the difference between hardware that works and hardware that fails is usually not talent or budget. It is method.
The problem with guessing
Complex hardware fails at the interactions. A connector can seal perfectly at room temperature and leak after a thousand heat cycles. A chamber component can hold its shape in isolation and warp once it is bolted into a system running at extreme temperature. The variables that drive these failures rarely act alone. They combine, and the number of combinations grows faster than any team can test one at a time. Change five design parameters with a few settings each and you already have hundreds of possible experiments. Testing them all is impossible. Guessing which ones matter is how good products fail in the field.
Nallagonda’s answer, refined across both industries, is design of experiments: a statistical approach that maps out a deliberately chosen subset of tests to reveal which variables matter and how they interact, in a fraction of the runs. Rather than change one factor at a time and hope, she structures experiments that vary several factors together, then reads the results for the hidden couplings that cause real-world failures. Paired with simulation, the method lets her rule out weak designs before any metal is cut, and aim for expensive physical testing only where it will actually teach her something.
“The instinct is to test everything, and that instinct quietly kills projects,” says Sahiti Nallagonda. “A well-designed experiment tells you where the real sensitivities are. You stop burning months chasing variables that do not matter and spend your time on the two or three that decide whether the part survives.”
When a wrong answer is an expensive answer
The stakes explain why this rigor matters. In semiconductor manufacturing, the margin for error has all but vanished. A single leading-edge fabrication plant now costs more than $20 billion to build, and the tools inside it are expected to run around the clock at near-perfect availability. When a hardware component inside one of those tools underperforms, the cost is not measured in the price of the part. It shows up as idle chambers, missed production ramps, and customers who cannot ship their own products on time.
That pressure has shaped Nallagonda’s semiconductor work. Leading the mechanical design of hardware for advanced etch processes, she helped introduce a first-of-a-kind architecture that wired electrical contact directly into a component that had previously relied on indirect coupling, opening up a new class of process control at the wafer edge. A change that fundamentally touches thermal behavior, mechanical stability, and electrical performance all at once, exactly the kind of tangled, multi-variable problem where intuition fails. Validating it took experiments designed to isolate each effect, so the new architecture could be trusted across nearly a thousand production chambers rather than a handful of prototypes.
“At that scale, it works on the bench means nothing,” Nallagonda notes. “You have to prove it holds across every chamber, every customer, every corner of the operating window. The only way to earn that confidence without testing forever is to design the experiments that matter and trust the data.”
Lessons from a different industry
Some of the sharpest lessons in hardware reliability come from the automotive world, where components ship by the tens of millions and a single failure mode, multiplied across a fleet, becomes a safety recall. Cars are unforgiving in a different way than chips. A part has to survive a decade of vibration, temperature swings from freezing to scorching, moisture, and road chemicals, all while costing pennies and assembling at high speed. That mix of extreme reliability and extreme cost pressure forces a discipline that transfers directly to any hardware problem.
Before her semiconductor work, Nallagonda led the design of a compact, high-current automotive connector built for the industry’s shift to smaller sealed interconnects, shrinking terminals to roughly half their former size without giving up current capacity or sealing performance. The hard problems lived in the physics of the contact itself: friction, wear, lubrication, and the force needed to mate the connector without tearing its seal. Working those couplings experimentally, she cut the force required to mate the connector by 38% while eliminating seal-tear failures outright, a result that became an internal benchmark. The connector went on to anchor a significant new revenue stream in sensing and engine applications.
“The connector taught me something I carried into semiconductors,” Nallagonda observes. “The failure is almost never where you first look. It is in some interaction between two things you were treating separately. Experiments designed properly are what surface those interactions before your customer does.”
The rising cost of getting it wrong
The demand for this kind of reliability is only intensifying, and nowhere faster than in the car. A modern battery-electric vehicle now carries roughly $1,200 of semiconductor content, close to triple a conventional car, spread across power modules, sensors, and the dense web of connectors and control hardware that ties them together. Each addition is one more part that has to endure years on the road, and one more interaction to master before the vehicle ships. The systems are getting more capable and less forgiving at the same time.
This is the environment Nallagonda has built a career around: hardware that has to be both extremely reliable and manufacturable at volume, where those two goals constantly pull against each other. Her approach treats manufacturing as part of the design problem from the start rather than an afterthought, using early experiments and simulation to find the tolerances and materials that will hold up in the lab, on the factory floor, and in the field alike. It is a way of working that closes the gap between an elegant design and a part that can be built a million times without drifting out of spec.
“Anyone can make one perfect part,” Nallagonda explains. “The real engineering is making the ten-millionth part as good as the first, in a factory, on a schedule. That is a statistics problem as much as a mechanical one, and most people learn it too late.”
Designing for what comes next
As chips push toward smaller geometries and cars toward full electrification, the hardware underneath is being asked to do more in less space, under harsher conditions, with less tolerance for error. The old rhythm of building, testing, and fixing in the field is becoming too slow and too costly to survive. The teams that keep up are the ones that treat experimentation as a design tool rather than a final exam, that learn where their designs are fragile early enough to do something about it. That shift, from testing to knowing, is quietly reshaping how serious hardware gets built.
Nallagonda’s career is a study in that shift. She has carried a single discipline across connectors and chambers, automotive and semiconductor, and made it portable: define the question precisely, design the smallest set of experiments that can answer it, then let the data rather than intuition choose the design. The specific hardware changes from one industry to the next. The method of getting it right does not.
“People think reliability is about being careful,” Nallagonda reflects. “It is really about being systematic. If you design the right experiments, the hardware tells you what it needs long before it reaches anyone who is counting on it.”
