ECE: Electrical & Computer Engineering
ECE News

Magnifying processing flaws
to make on-chip fingerprints

Students at a computer looking at chips on the desk before them

Dinesh Ganta, chief tester; Mike Henry, chief designer, and Lalleah Rafeei test chips for identifying fingerprints and function. Leyla Nazhandali and Patrick Schaumont encourage undergraduates to join their research teams. Rafeei was an undergraduate researcher who then chose to pursue a graduate degree.

Random variations in chip manufacturing can be exploited to create unique, stable on-chip identification, according to a research team at Virginia Tech. ECE’s Leyla Nazhandali and Patrick Schaumont, working with Inyoung Kim of statistics, have developed techniques to amplify naturally occurring manufacturing variations — without hurting chip functions.

Current chip identifiers are simple pieces of data that can be easily copied just like the social security number of an individual. “We would like to build chip identifiers that are like human fingerprints in nature, so that they cannot be copied” Nazhandali says.

Process manufacturing variations are caused by the limitations of photolithography, explains Nazhandali. “Because of that and non-uniform conditions during fabrication, no two chips are identical. They all have slight differences,” she says.

Their team is using those typically undesirable traits to make unique, trustworthy electronic hardware fingerprints. They are magnifying the process variations to make it stable in spite of temperature changes that occur during chip operations. “We are going against the typical paradigm,” says Nazhandali. “Manufacturers try to reduce process variation and we want to magnify it.”

Leyla Nazhandali holds a chip up to the camera

Leyla Nazhandali displays an ASIC chip manufactured to amplify processing flaws without impacting function. The flaws can be used as unique chip identifiers for secure applications.

Using special circuits and techniques, the team has fabricated prototype ASIC chips with a magnified process variation effect. “Our results to date show we can improve the stability of the chip identification by a wide margin — about 18 percent,” she reports. “Yet the chips still function to specifications.”