Analysis of Transistors During Operation: A Great Debugging Tool and a Concrete Security Risk
A group of researchers from the University of Adelaide has developed an experimental technique that allows for the analysis of a chip's internal transistor activity while it is operational, utilizing radiation in the terahertz band. The result paves the way for a new measurement method that could significantly improve the validation and diagnostic processes of semiconductors.
As reported by IEEE Spectrum, the system is based on a vector network analyzer (VNA), a tool capable of generating microwave signals with known frequency and phase. A frequency extender converts this signal into terahertz waves, which are directed towards the microchip through a focusing lens. The test requires the chip to be powered on and operational: in this condition, the active transistors modify the reflected signal.
The returning signal is then reconverted back to the microwave band and compared with the original. The detection of minimal differences in amplitude and phase occurs via a quadrature homodyne receiver, a fundamental component for obtaining reliable measurements. The researchers had to adapt the hardware of the receiver, originally designed to work only with microwaves, to operate in the terahertz domain.
From a physical standpoint, an important criticality emerges: the wavelength of the terahertz signal is greater than the dimensions of the observed transistors. This aspect complicates the identification of variations in the reflected signal. Additionally, the noise generated by the VNA's oscillator could mask differences, necessitating extremely high sensitivity in the measurement chain.
The main advantage of the technology lies in the ability to observe the internal behavior of a processor during processing, a feature not found in current tools. This approach opens interesting scenarios for advanced debugging and functional analysis of chips.
However, there are limitations. Modern chips, especially those with 3D stacked architectures and multilevel chiplets, present a complex structure that hinders selective reading of the various layers. Terahertz waves cannot accurately distinguish the level from which the signal originates when the upper layers are opaque. In this regard, the researchers are exploring techniques to increase the sensitivity of the VNA and improve resolution in these scenarios.
On the security front, a potential risk emerges: prospectively, the same technique could be utilized to intercept information during the processor's operation. Encryption systems would not be sufficient to counter this type of analysis, as the data must be decrypted before processing. While this remains a theoretical hypothesis, it could necessitate dedicated countermeasures when the technology reaches commercial maturity.