NTT achieves world first in evaluating energy consumption of information processing circuits through thermodynamic

NTT has successfully become the world's first to simultaneously measure heat and entropy (information entropy) on a single-electron scale. They achieved this by applying a single-electron detection technology that utilizes an original nanometer-scale electronic device capable of detecting changes in the number of electrons one by one, within a semiconductor memory element (DRAM cell) operating at room temperature.

Through this breakthrough, the researchers experimentally verified the theoretical limit of minimum energy consumption in information processing, a feat previously considered impossible. They revealed that the reason energy consumption exceeds this theoretical limit is not only due to high-speed processing and the influence of peripheral circuits, as traditionally pointed out, but also due to thermal instability when retaining information.

This achievement opens the path to evaluating the energy consumption of information processing circuits from a thermodynamic perspective. It is highly expected to be applied to the development of information processing devices with superior energy-saving capabilities and next-generation memory technologies.

In information processing, an "initialization operation" is performed to align scattered information into a fixed state. During this process, entropy, which represents the randomness of information, decreases, and heat generation (energy consumption) occurs as a trade-off.

The theoretical minimum value of this energy consumption is known as the Landauer limit, which serves as a crucial benchmark when designing energy-saving information processing.

However, actual electronic devices consume energy that significantly exceeds this theoretical limit, and uncovering the cause of this gap has been a major challenge in creating energy-saving devices. Furthermore, while it was conventionally believed that this discrepancy was caused by the effects of high-speed processing and peripheral circuits, there had been no experimental verification.

Therefore, NTT aimed to clarify this energy-saving limit in their latest study. To eliminate the influences of high-speed processing and peripheral circuits, they focused on the circuit structure of a DRAM cell, which is the smallest unit of a memory element that constitutes one bit in DRAM, and verified whether it could reach the Landauer limit when the initialization operation was performed at a slow speed.

However, under extreme conditions approaching the Landauer limit, the entropy and heat signals required for verification are extremely minute and easily buried in noise. There was previously no method available to measure semiconductor elements operating at room temperature. As a solution to this problem, NTT applied single-electron detection technology using its proprietary silicon nanodevices.

Utilizing a high-performance detector fabricated with microfabrication technology, they successfully measured the amount of electric charge stored in a capacitor on a single-electron basis, and evaluated heat and entropy based on that data.

When electrons move between a lead and a capacitor during information processing, heat is generated or absorbed according to the potential difference between the two. Therefore, if the electrical potentials of the lead and the capacitor at the exact moment of movement are known, the amount of heat can be calculated.

In this study, the potential of the capacitor was calculated from the amount of electric charge, and heat was measured by combining this with the potential of the lead, which is already known as an externally applied voltage. In addition, because information can be accurately obtained by measuring the amount of charge in the capacitor, which determines the information of the DRAM cell, on a single-electron scale, the researchers were also able to calculate entropy.

In this DRAM cell, one bit of information is recorded according to the amount of charge stored in the capacitor. Increasing the amount of charge injected during initialization reduces the initialization error rate caused by thermal fluctuations, which in turn increases the amount of entropy reduction.

In the experiment, the researchers varied the amount of injected charge under multiple conditions to investigate the relationship between the amount of entropy reduction and the amount of heat.

The results showed that as entropy decreased, the amount of generated heat increased, and the rate of increase grew larger as well. This indicates that as the initialization error rate decreases, the amount of heat deviates further from the Landauer limit.

In other words, the cause of this discrepancy cannot be explained solely by the effects of high-speed processing and peripheral circuits, yielding an important new insight that overturns conventional understanding.

Furthermore, detailed analysis revealed that because the DRAM cell holds information in a thermally unstable state (non-equilibrium state), extra heat is generated when it transitions to a thermally stable state (equilibrium state) midway through the initialization operation, making it impossible to achieve the Landauer limit.

Based on the findings above, it was determined that to create energy-saving memory elements that approach the Landauer limit, it is necessary to utilize a structure capable of holding information in a thermally stable state.