VO₂ memristor-based frequency converter with in-situ synthesize and mix for wireless internet-of-things

Wireless IoT (WIoT) systems span applications such as smart homes, industrial automation, and medical monitoring, often requiring long-range links at low frequency bands down to tens of kHz. Conventional frequency converters for such links are built with CMOS digital and analog circuits that separate frequency synthesis (registers, oscillators, DACs) and mixing (op-amp based mixers), targeting unnecessarily high frequencies (up to GHz) and thereby incurring excessive latency and energy consumption for low-frequency WIoT. A compact frequency converter that can synthesize and mix within a single module at up to tens of kHz is highly desirable. Memristors, with rich ion dynamics and tunable oscillatory behavior, offer a promising path: while oscillatory bionic neurons have been reported, no work has leveraged their oscillation for frequency conversion or demonstrated a calibratable memristor oscillator array performing in-situ synthesize-and-mix in an end-to-end WIoT setup. This work proposes and demonstrates a VO₂ memristor-based oscillator array that exploits negative differential resistance to realize in-situ frequency synthesis and mixing, aiming to reduce latency and power while meeting WIoT low-frequency needs.

The paper reviews state-of-the-art CMOS frequency converters employing separate synthesizer and mixer architectures with frequency control registers, parallel oscillators, DACs, and op-amp mixers, supporting up to GHz operation and leading to inefficiency for WIoT low-frequency bands. Prior works have developed high-frequency CMOS synthesizers and cryo-CMOS systems, but these are over-provisioned for kHz-range WIoT. In the memristor domain, extensive research has explored oscillatory devices and neuromorphic functions (e.g., VO₂ and other Mott memristors, phase-change neurons, diffusive memristors), yet no prior literature utilizes memristor oscillations specifically to build a frequency converter, nor reports a calibratable oscillator array enabling in-situ synthesize-and-mix with a full WIoT demonstration. This gap motivates the proposed VO₂ memristor-based approach.

Device fabrication and characterization: 20 nm VO₂ films were epitaxially grown on c-plane Al₂O₃ substrates via pulsed laser deposition (248 nm KrF excimer laser, ~2 J cm⁻² energy density, 5 Hz, 500 °C, 1.0 Pa O₂). Films were cooled at 20 °C min⁻¹. Devices use a lateral Au/Ti (50/5 nm) planar geometry with 5 µm channel length and 5 µm electrode width patterned by EBL, evaporation, and lift-off. Material quality was confirmed by electrical MIT (~400× resistance change; critical temperatures ~345 K heating, ~339 K cooling), XRD showing monoclinic (020) VO₂ on Al₂O₃, Raman spectra aligned with monoclinic VO₂, XPS indicating appropriate V oxidation states (V⁴⁺, V⁵⁺ surface contribution), and TEM/STEM-EDS revealing high crystalline quality. Oscillator design and modeling: VO₂ memristors exhibit negative differential resistance enabling self-oscillation under current bias. A simple oscillator connects a VO₂ device in parallel with a capacitor; a 25 Ω resistor (R₀) converts current to an observable voltage output. Oscillations are induced with bias currents in the NDR region and measured on an oscilloscope. Frequency is modulated by input current and parallel capacitance. An electro-thermal SPICE compact model accurately reproduces I–V characteristics and oscillation behavior. Devices showed endurance >10⁶ cycles. Frequency increases with temperature; temperature dependencies should be modeled in deployment. Calibration and array architecture: To address device-to-device variation across a 1 cm² chip, a parallel calibration resistor (R_c/R_e) is added to each oscillator. Adjusting R_c tunes the firing frequency and the f–I_in curve, substantially reducing D2D variation and improving uniformity while epitaxial growth minimizes cycle-to-cycle variation. An 8×8 VO₂ array (eight 8×1 rows) was fabricated; one electrode of each device is directly driven by a current source and the other electrodes are interconnected to accumulate outputs. Frequency converter operation (in-situ synthesize and mix): The array enables summation (accumulation) of pulse outputs for same or different frequencies by simultaneously biasing multiple oscillators. Experiments demonstrate: two-channel same-frequency accumulation (e.g., 1.41 mA and 1.42 mA biases), different-frequency accumulation (e.g., 1.5 mA and 1.22 mA biases), subtraction via simultaneous positive/negative current biases, and eight-channel accumulation (eight distinct current biases). The effective V_out matches the sum of individual channels. Frequencies are programmable via array size, current-driver settings (voltage-controlled current mirrors), and calibration resistances. Supported frequencies reach up to 48 kHz. End-to-end WIoT system: A hardware–software co-designed setup encodes sensor data using Huffman source coding followed by rate-1/2 convolutional coding and BPSK modulation (40 samples/bit). Control voltages computed in software drive on-board CMOS current drivers feeding the VO₂ array to synthesize and mix frequencies in situ. Due to non-sinusoidal oscillator waveforms and potential phase misalignment, programmable delay-line phase matchers (LTC6994) align phases prior to RF modules (filter, attenuator, power amplifier). Signals are transmitted over an AWGN channel with varying SNR and received data are demodulated and decoded. Three representative sensor datasets are transmitted: audio (3 s, Fs = 4.41 kHz, 52.82 kB), image (300×300 RGB, 270 kB), and point cloud (XYZ, 176.946 kB).

• VO₂ memristor oscillators leveraging NDR achieve stable, programmable self-oscillations. Frequency increases with input current and decreases with added parallel capacitance; SPICE electro-thermal modeling matches experiments. • High endurance (>10⁶ cycles) and high crystalline quality yield low cycle-to-cycle variation. Device-to-device variation is effectively mitigated via parallel calibration resistors (R_c), aligning f–I_in characteristics across oscillators. • An 8×8 VO₂ array enables in-situ frequency synthesis and mixing for 2–8 channels with frequencies up to 48 kHz, performing same-frequency summation, different-frequency summation, and subtraction (via positive/negative biasing). The combined V_out equals the sum of individual channel outputs. • End-to-end WIoT demonstrations (audio, vision, spatial) show that the VO₂-based frequency converter incurs minimal BER degradation compared with CMOS-based FC: additional SNR required at BER = 10⁻⁴ is approximately 0.02 dB (audio), 0.13 dB (vision), and 0.21 dB (spatial). • Power benefits: VO₂-based FC achieves 1.45×–1.94× lower power consumption than conventional CMOS FC in the tested low-frequency regime while maintaining functional performance. • Programmability and calibration through array size, voltage-controlled current drivers, and calibration resistances enable flexible frequency assignment within the target low-frequency bands.

The proposed VO₂ memristor-based oscillator array directly addresses the inefficiency of conventional CMOS frequency converters in low-frequency WIoT links by unifying synthesis and mixing within a single, compact module. The NDR-enabled oscillators provide tunable frequencies that, when summed across channels in the array, realize in-situ frequency mixing without separate DACs and op-amp mixers. Calibration resistors effectively standardize oscillator behavior across devices, improving array-level uniformity and practicality. The end-to-end WIoT experiments confirm that despite non-sinusoidal waveforms, phase-matching can enable integration with RF modules, yielding negligible BER penalties relative to CMOS FC while delivering significant power savings. These results validate the feasibility and benefits of a memristive frequency converter for kHz-range WIoT transmission, highlighting reduced latency/energy and programmable operation. Practical deployment considerations include temperature-dependent frequency shifts and the need for phase alignment, both manageable with modeling and compact periphery circuits.

This work demonstrates a VO₂ memristor-based frequency converter that performs in-situ frequency synthesis and mixing using a calibratable 8×8 oscillator array. The epitaxially grown VO₂ devices produce robust, tunable oscillations with high endurance and uniformity enhanced by parallel calibration resistors. The converter supports 2–8 channel mixing up to 48 kHz and, in an end-to-end WIoT setup, achieves 1.45×–1.94× power improvement with only 0.02–0.21 dB BER degradation versus a CMOS-based baseline. These results establish memristive oscillator arrays as a promising solution for low-frequency WIoT systems, offering reduced complexity, latency, and energy. Future efforts can focus on broader frequency scaling, tighter integration of periphery circuits (e.g., on-chip phase alignment), and temperature-aware control to further enhance robustness and scalability for real-world deployments.

The demonstrated converter targets low-frequency bands (up to 48 kHz), which may limit applicability to higher-band WIoT or RF systems. Oscillator outputs are not perfectly sinusoidal, necessitating external phase matching to meet RF module requirements. Device-to-device variation exists and requires calibration resistors to standardize behavior. Oscillation frequency is temperature dependent, implying a need for temperature-aware models and compensation in practical systems. The prototype relies on external CMOS-based current drivers and delay-line phase matchers, indicating additional integration work for a fully on-chip solution.