As Moore’s Law continues dictating the pace of technology innovation, it’s driving computer chip engineers into an uncharted world that is exponentially shrinking and extremely complex.
For nearly two decades, the pursuit of perfection has led Nikos Troullinos down minuscular rabbit holes to fix tiny design mistakes that can cause computer processor circuitry to malfunction.
The Intel chip engineer’s calm demeanor masks the magnitude of complexity he manages every day, eliminating imperfections trapped among the billions of interconnected objects that fit on a sliver of silicon, often smaller than a thumbnail.
“It’s like cutting a hole in the sky, plucking out that piece to remove an erratic star, then replacing that piece back into the sky without disrupting the cosmos,” said Troullinos, describing how he and his team repair designs before computers chips are mass manufactured.
The astrophysical analogy is apt when you consider zooming to 14 billionths of a meter, the size of transistors that power the latest 5th generation Intel Core processors.
“We take a block that has a billion objects on it then focus on one thousand, which is like zeroing in a million times,” he said.
Intel Senior Fellow Mark Bohr illustrated in a recent video what it means to be in a space surrounded by transistors measuring 14 nanometers.
Today, whenever Troullinos goes to work on a future chip design, he operates in an infinitesimal universe where transistors measure 10 and even 7 nanometers. He points out that the diamond lattice distance of silicon is 0.54 nanometers. To put that into perspective, the diameter of a human hair is about 75 microns, which equals 75,000 nanometers.
“We often work on designs that are two generations ahead of what’s being manufactured,” he said.
To find and fix faulty transistors and layers of interlacing connectors requires sophisticated software tools that evolve dramatically in order to keep up with Moore’s Law, which dictates the number of transistors will double on the same size of silicon every two years.
This happens because engineers keep finding ways to make smaller, more efficient transistors and wires to connect them up, but Troullinos, like many others in his field, see that this is getting more difficult to do.
“Engineering is the art and science of solving hard problems and workarounds for exceptionally difficult problems to reach near perfection,” said Troullinos, who joined Intel in 1996.
He studied electrical engineering at Aristoteleion Panepistimion of Thessaloniki in Greece before earning a PhD in computer science at Syracuse. Around Intel, he’s known for helping design software tools that speed the chip design process.
“Software tools are as important as the technology being created,” he said. “Software tools enable our folks to design and make these incredibly complex chips.”
He said the tools aren’t perfect, but they keep evolving and improving, because they have to keep up with Moore’s Law.
While science-gurus-turned-TV-stars Adam Savage and Jamie Hyneman satirized Moore’s Law in a 2007 video for Intel, the nano science driving Moore’s Law into the future is increasingly tedious and complicated.
Troullinos said that not only does he have to deal with smaller circuitry every two years, he also has to manage multiple complexities borne from integrating more intricate analogue and digital functions, not to mention the fusing of central processing units (CPU) with graphics or LTE (4th generation wireless communication) like the Intel Atom x3 system on a chip, codenamed SoFIA.
“The real world is analog and we created digital to simplify or make things repeatable,” he said. “Analog is gradual and continual while digital is limited to specific values like how binary is limited to two: on or off, yes or no.”
Before any chip design moves into mass manufacturing it needs to meet strict quality standards. The “light pen” used to pattern today’s chips is 15 times wider than the slimmest circuitry. This leads to many tradeoffs and is the reason his team made the software tools, which make the repairs faster.
He describes it as trying to fix a problem in a tight, dark place. First the erratic area needs to identified, enlarged and lit before surgical repairs can be made.
“It used to be done by hand, but increasingly we are training computers to do it,” he said.
“After a dozen times of fixing a small portion of a chip, we can program or train machines to do the fixes.”
These images show progress of recent repairs on a piece of an upcoming 10 nanometer chip. This block is about 50 times smaller than the actual chip. Thousands of tiny repairs were done in a day with the help of engineers in the U.S., Costa Rica and Penang working in parallel.
“What used to take several days or weeks to clean up now can be done in a day,” said Troullinos.
Without these software tools, it’d be tough to keep Moore’s Law alive, he said.
At the edge of perfection Troullinos finds the need for variation, something he sees as essential to life and electronics.
“Inside circuitry, aspects of change are what make it function. It’s these variations that bring them to life.”
Editor’s note: This video shows Dave Pivin, an analysis engineer at Intel who looks for defects in chips once they’ve been manufactured on silicon wafers. “My job is like finding a needle in a haystack, and I’m looking in a million haystacks for a single needle,” he said. “These days at 22 nanometer and 14 nanometers, the defects are so small they’re almost invisible.”