Proposal for a Low Voltage Analog-to-Digital Converter using Voltage Controlled Stochastic Switching of Low Barrier Nanomagnets

Indranil Chakraborty, Amogh Agrawal, Kaushik RoyarXiv 2018 · arXiv

Researchers at Purdue University have proposed a novel approach to building Analog-to-Digital Converters (ADCs) by embracing the inherent "noise" of nanoscale physics. By leveraging the stochastic switching of low-barrier nanomagnets controlled by voltage, the team presents a design that could significantly reduce the power and area footprint of ADCs in neuromorphic and Internet of Things (IoT) applications.

The Challenge of Traditional ADCs

Traditional Analog-to-Digital Converters, particularly Flash ADCs, rely on large banks of comparators. As resolution increases, the number of comparators grows exponentially ($2^n - 1$), leading to high power consumption and large silicon area. This scaling issue is a major bottleneck for "edge" devices like sensors and neuromorphic chips that require high efficiency. While current-driven spintronic devices have been proposed as alternatives, they often suffer from ohmic losses. This research addresses the need for a voltage-controlled spintronic ADC that operates at ultra-low power while maintaining high precision.

The Magneto-electric Magnetic Tunnel Junction (ME-MTJ)

The core of the proposed design is the Magneto-electric Magnetic Tunnel Junction (ME-MTJ). Unlike traditional magnetic memory which seeks to maximize stability, this device utilizes "low-barrier" nanomagnets where the energy barrier ($E_B$) between magnetic states is comparable to thermal energy (~$1kT$).

Structure of the ME-MTJ device showing the layers and read/write paths.

As shown in Figure 1, the device consists of a Free Layer (FL) and a Pinned Layer (PL) separated by a tunnel barrier (MgO), sitting atop a Magneto-electric (ME) Oxide. By applying a small analog voltage to the oxide layer, the system can influence the magnetic state of the FL via the ME effect, changing the probability of the magnet being in one state versus the other without the need for high-current injection.

Stochasticity as a Resource

In this design, thermal instability is treated as a resource rather than a nuisance. At $E_B \approx 1kT$, the magnet becomes "superparamagnetic," flipping spontaneously between states at gigahertz rates due to ambient heat. This thermal noise provides a built-in "dither" that helps linearize the conversion process.

Comparison of stable (40kT) and stochastic (1kT) switching dynamics and the resulting linear response to voltage.

As illustrated in Figure 3, while a high-barrier magnet ($40kT$) remains stable, the low-barrier magnet ($1kT$) exhibits rapid telegraphic switching. Crucially, the time-averaged magnetization ($<m_x>$) follows a sigmoid curve relative to the input voltage. Within a specific range (-0.4V to 0.4V), this response is highly linear, allowing the device to act as a natural quantizer.

Circuit Implementation and Results

The ADC system integrates the ME-MTJ with a sensing circuit and a ripple counter. The sensing circuit, comprising a voltage divider and CMOS inverters, converts the magnet's instantaneous resistance into a digital '0' or '1'. The ripple counter then tracks these states over a sampling window ($t_s$). The final count becomes the digital representation of the analog input.

Key performance metrics from the study include:

  • Precision: Achieved up to 8-bit precision.
  • Power Consumption: Worst-case power consumption was calculated at 34 $\mu$W for the sensing and counting logic.
  • Linearity: The design showed a Normalized Root Mean Square Deviation (NRMSD) of approximately 2% for 8-bit precision.
  • Speed: An 8-bit conversion requires a sampling time of 10 $\mu$s.

Limitations and Future Outlook

While promising, the design faces a fundamental trade-off between precision and speed; higher resolution requires longer sampling of the stochastic flips, which limits bandwidth. Additionally, because the energy barrier is so low, the device is highly sensitive to temperature fluctuations. Future work will need to address material challenges in fabricating ME-MTJs with precisely controlled low energy barriers and maintaining stability across different environments. This design is particularly well-suited for "approximate computing" paradigms where ultra-low power consumption is prioritized over absolute bit-perfection.

Summary updated

BibTeX
@misc{chakraborty2018proposal,
  author = {Indranil Chakraborty and Amogh Agrawal and Kaushik Roy},
  title = {{Proposal for a Low Voltage Analog-to-Digital Converter using Voltage Controlled Stochastic Switching of Low Barrier Nanomagnets}},
  year = {2018},
  eprint = {1803.01431},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1803.01431},
}
Figure