Multichannel power electronics and magnetic nanoparticles for selective thermal magnetogenetics

We employ series L–C oscillation to drive alternating current in a high-flux petri dish coil (L_coil = 3.14 μH, see section 2.2) to generate the alternating magnetic field. Three capacitor banks are attached to the same coil, with each bank tuned to achieve a single channel frequency. This design allows simple expansion to include additional channels. A hybrid full-bridge circuit generates rectangular voltage output with adjustable duty cycle and frequency controlled by a system-on-chip (SoC) controller. The duty cycle controls the coil current's amplitude (pulse width modulation), and the frequency is tuned to match the resonance frequency of one of the L–C banks to excite it. The series resonance as a sharp notch-filter presents a large impedance to all frequency contents of the rectangular driving voltage except for its fundamental so that a sinusoidal current follows. Once tuned, the driving circuit only needs to operate at a low voltage (here ±36 V peak for the rectangular waveform) and the resonance design shields off the reactive component. Despite the low voltage on the electronics side, the coil and capacitor voltages can reach 1 kV rms. On the other hand, the coil current (up to 500 A rms for Channel 1—Ch 1) must entirely flow through the circuit, requiring proper wiring and cooling design.

Instead of one transistor technology for the entire range of frequencies and currents, two specialized device types, each optimized for a part of the frequency range, work collaboratively for various trade-offs (figure 1): for Channel 3 (Ch 3), the power switches of the circuit need to operate on the order of f ⩾ 2 MHz but only conduct low current (e.g. I_coil ⩽ 10 A rms). Latest high-electron-mobility GaN-type transistors appear to fit this scenario perfectly, due to their lower parasitic capacitance at gate and output, shorter carrier lifetimes, higher critical field strength, and higher charge carrier mobility [24, 25]. In contrast, the operating current of Ch 1 can reach beyond 500 A. Vertical Si power trench FETs are more suitable here due to their large die areas. The electronic circuit is modularized, with each module sharing a similar layout for better scaling. The full-bridge circuit for the 50 kHz channel is implemented with 16 Si FETs (IPT015N10N5, Infineon Technologies AG, Neubiberg, Germany) for each switch due to the peak current demand. At 500 kHz, the controller activates four FETs per switch at a time in a round-robin fashion to reduce the switching loss. As for the MHz channel, the driving voltage is provided by a full bridge with four GaN transistors (GS61008T, GaN Systems, Ottawa, Canada).

Zoom In Zoom Out Reset image size

We use contactors to decouple different channels. The contactors (B88269X1090C011, TDK Electronics AG, Munich, Germany) avoid circulating currents between different channels and allow the magnetic coil to be electrically connected to one set of the corresponding series capacitors for each channel at a time. The contactors are driven by lower power FETs (ZVN4210A, Diodes Incorporated, Plano, TX, USA) via a SoC controller. Supplementary table S1 (available online at stacks.iop.org/JNE/19/026015/mmedia) lists the major electronic components and technical specification of the three channels.

The capacitor bank is implemented with polypropylene film capacitors for high voltage, high current, and high frequency applications (Cornell Dubilier Electronics, Liberty, SC, USA and WIMA GmbH & Co. KG, Mannheim, Germany). Quick-change sockets (SparkFun Electronics, Niwot, CO, USA) are used so that capacitors can be switched in and out to tune the frequency of the resonance circuit. As the power electronics' output frequency is limited by the clock frequency (100 MHz) of the SoC, the adjustment in the megahertz range is more granular and the output frequency of Ch 3 might not meet the exact resonance frequency of its L–C circuit using discrete capacitors. Therefore, a variable capacitor (10–100 pF, Sprague Goodman Electronics, Westbury, NY, USA) operates in series with the replaceable film capacitors, which allows the user to tune the circuit to resonate at the discrete frequencies output in the megahertz range and maximize the output current and magnetic field. For Ch 1 and Ch 2, the frequency step size is fine enough. Thus, resonance can be reached by adjusting the output frequency via the user interface. The voltage rating of the resonating capacitors should be high enough to accommodate the peak operating voltage up to 1 kV across the capacitors and inductors.

A SoC with field-programmable gate array (FPGA) and ARM Cortex A9 microcontroller (Xilinx Zynq 7000, sbRIO-9627, National Instrument, Austin, TX, USA) controls the transistor banks and the contactors. The SoC connects to a PC with a graphic user interface (LabVIEW, National Instruments, Austin, TX, USA) through an ethernet–USB adapter, which selects and sets the stimulation channels (pulse duration, interpulse interval, stimulation frequency, and amplitude, etc). The FPGA part of the SoC performs all transistor switching, for which we clocked it at 100 MHz to refine the time resolution so that the driver circuit can hit all necessary frequencies close enough.

The electrical cable of approximately 30 cm length consists of six interleaved high-voltage insulated 12-gauge (American wire gauge) type-8 Litz wires (New England Wires Technologies, Lisbon, NH, USA), with three wires in each direction. The cable is ensheathed in interference-shielding heat-shrink tubing to reduce radio-frequency noise leakage and terminated in compression lugs that connect to the coil.

All electronics components were placed in a standard four-height-unit 19 inch vented rack (Hammond Manufacturing, Guelph, Canada), with the aluminum enclosure acting as shielding to limit RF noise leaking to the nearby equipment. Heat sinks were added on the electronics boards to increase thermal capacity and surface area, and each circuit module had two fans, one above and one below, for cooling. A 12 V AC/DC adapter powers the FPGA and fans and five 36 V 1 kW DC power supplies (Mean Well USA, Fremont, CA, USA) powered the electronics. A layout of the integrated system is shown in figure S1.

The magnetic field coil (Hi-Flux module, MSI Automation, Wichita, KS, USA) consists of a six-turn solenoidal coil of 5.75 cm inner diameter and 3.8 cm height wound with a 3/16 inch (4.76 mm) diameter copper tube, which allows liquid cooling with water (figure S2). The coil contains a ferrite core (μ = 2300, Ferroxcube 3C90, Eindhoven, Netherland) in a sealed compartment. The coil is located with its cylindrical axis vertically within a nylon panel enclosure on which petri dishes can be placed for microscope imaging. The coil and core were potted in high-temperature epoxy (Devcon, Milpitas, CA). The upper rim of coil windings and ferrite inside the compartment are 3 mm below the surface for placing petri dishes. The coil is electrically connected via two ¼-20'' bolts to the cable, and both the coil and the ferrite compartment are cooled via chilled circulation attached via metal quick-disconnect fittings (CPC Colder, St. Paul, MN, USA). The coil's electrical characteristics were measured in four-wire configuration with a bench LCR/ESR meter (Model 889A, B&K Precision Corporation, USA).

We performed all electrical measurements with a 500 MHz high-bandwidth oscilloscope (Tektronix MDO3054, Beaverton, OR, USA). High-voltage differential probes (Tektronix THDP0100) served to measure the coil voltage, whereas Rogowski current transducers (CWT 60B & CWT MiniHF 1B, Power Electronic Measurement Ltd, Nottingham, UK, for Ch 1 & Ch 2 and Ch 2 & Ch 3, respectively) detected the coil current. To measure the magnetic field generated by the coil, we used a magnetic flux test coil of ten turns by winding 30-gauge magnetic wire around a 3D-printed polylactic acid cylindrical former of 1 cm diameter (Ultimaker B.V., Utrecht, Netherlands). The probe measures the voltage induced by the changing magnetic flux of the alternating magnetic field and the field strength was derived in either the time or frequency domain.

Resonance frequencies for each channel were confirmed by varying the output frequency of the electronics circuit and maximizing the measured coil current/voltage or induced voltage of the magnetic field probe. Once the resonance frequency was recorded, it was not changed any more during operation.

We tested sequential cycling through the three channels with the following settings: Ch 1, 49.9 kHz, 90 mT; Ch 2, 549 kHz, 13.7 mT; Ch 3, 2.5 MHz, 3.9 mT (see results on frequency and amplitude combinations of the channels). The channels were cycled in the order from lowest to highest frequency with equal pulse duration of 3 s and two inter-pulse intervals of 17 s and 7 s were tested, corresponding to a duty cycle and coil current of 15% with 88 A and 30% with 125 A, respectively. Temperatures of the coil, cable, and electrical and electronics components within the rack were measured using a handheld infrared thermometer (Fluke Corporation, Everett, WA, USA).

Data analysis was performed in MATLAB (R2019b, The MathWorks, Natick, MA, USA). To calculate the magnetic field, the induced voltage was detrended to remove any DC offset, filtered above 50 MHz, and integrated.

Three types of magnetic nanoparticles with different coercivity were developed to form three magnetothermal channels. Our 15 nm cobalt-doped iron oxide (Co_0.65Fe_2.35O₄) nanoparticles generate a large amount of heat in a low-frequency high-amplitude magnetic field (Ch 1), 19 nm iron oxide (Fe₃O₄) nanoparticles in a medium-frequency medium-amplitude field (Ch 2), and 10 nm manganese-doped iron oxide (Mn_0.57Fe_2.43O₄) nanoparticles in a high-frequency low-amplitude field (Ch 3). The synthesis, coating, and functionalization of these nanoparticles followed previous methods [18, 26]. Briefly, magnetite nanocrystals of 4 nm diameter were first synthesized by thermal decomposition of iron acetylacetonate in a mixture of oleic acid and benzyl ether and increased to 15 nm diameter size by controllable seed-mediated growth in a mixture of iron acetylacetonate (Fe(C₅H₇O₂)₃), oleic acid, and benzyl ether. Then, the nanocrystals were coated with a copolymers of phospholipids and polyethylene glycol (DSPE-PEG2K) using a dual-solvent exchange method [27]. The cobalt-doped nanoparticles were synthesized from multiple seed-mediated growth reactions through thermal decomposition with 5 nm iron oxide cores in 2 mmol CoCl₂, 4 mmol Fe(C₅H₇O₂)_3, 25 mmol oleic acid, using 60 ml benzyl ether as a solvent. Manganese-doped seeds of 7 nm size were synthesized with a mixture of 5 mmol MnCl₂, 10 mmol Fe(C₅H₇O₂)₃, 50 mmol 1,2-tetradecanediol, 60 mmol oleic acid, 60 mmol oleylamine and 60 ml benzyl ether. The seeds were then grown to 10 nm with a seed-mediated growth procedure. The reacting solution was heated to 120 °C for 30 min under a constant argon flow, then to 200 °C for 2 h and finally to reflux at 300 °C for 30 min. The product was purified through several acetone washes. The nanoparticle sizes were determined by high contrast transmission electron microscopy (TEM). Nanoparticles were then coated with DSPE-PEG2K by mixing the nanoparticles with PEG and adding dimethyl sulfoxide (DMSO) . The reaction was then evaporated and transferred to water by a drop-wise addition of water and removal of the remaining DMSO by was done by centrifugation and ultracentrifugation.

TEM, superconducting quantum interference device (SQUID), and power x-ray diffraction (XRD) measurements were performed to quantify the size distribution, magnetic properties, and the crystal structure of the magnetite nanocrystals, respectively. The hydrodynamic size of conjugated nanoparticles was subsequently examined by dynamic light scattering. The samples for TEM measurements were prepared by diluting the samples and placing them in carbon-film grids. XRD samples were prepared by drying the nanoparticles under an argon flow and then pulverizing the resulting powder. SQUID measurements were performed with coated samples by fixing the nanoparticles with calcium hemisulfate and enclosing them within a capsule to prevent movement. Doping percentages were determined by inductively coupled plasma mass spectrometry and comparing the samples to the corresponding standard curves of iron, cobalt, and manganese.

The nanoparticle samples were placed on the surface of the coil and the heating was measured using an infrared thermal probe (Luxtron 812 and STF-2 M probe, Lumasense, Santa Clara, CA, USA). The specific absorption rate (SAR) is calculated as

$$\begin{align} {\text{SAR}} = \frac{C}{\rho } \cdot \frac{{{{\Delta }}T}}{{{{\Delta }}t}}, \end{align}$$

where $C$ is the specific heat capacity of the media (4180 J kg⁻¹ K⁻¹), ${{\Delta }}T$ is the temperature change during stimulation averaged over three stimulations, ${{\Delta }}t$ is the stimulation time, and $\rho$ is sample density measured by total metal concentration. The undoped, Co-doped, and Mn-doped iron oxide nanoparticles were recorded at 4.33 mg_metal ml⁻¹, 9.58 mg_metal ml⁻¹, and 28.18 mg_metal ml⁻¹, respectively. Thermal images were recorded with an infrared camera (FLIR A700, FLIR, Wilsonville, OR, USA).

We characterized the power electronic system driving a high-flux coil to determine whether their performances meet the design goals for heating the nanoparticles. A relative uniform distribution of magnetic field over the central 1 cm region within the petri dish chamber of the coil (figures S2 and S3) ensures that several nanoparticle samples can be exposed to the required field strength when placed above the coil.

While the software ensures hitting the resonant frequencies, the reconfigurable capacitor arrays allow further adjustment of the resonant frequencies. Ch 1 resonates at 49.7 kHz and 61.5 kHz with 2 µF and 3 μF capacitors, respectively. Ch 2's output offers 46 frequencies in the 380–580 kHz frequency range as allowed by the system's clock, and the system was characterized with approximately 25 kHz steps between 400 and 575 kHz. Ch 3 had 15 frequencies within the 2–5 MHz range allowable by the system's clock. The output frequency for each channel can be adjusted using the capacitor combinations in table S2.

The measured magnetic field amplitude (figure 2) shows that the target field strength can be achieved for each channel. However, due to the higher impedance (figure S4) and losses with increased frequency, the achievable magnetic field decreases for higher frequencies and the envisioned target of 1.25 mT was only achieved for frequencies up to ∼4 MHz.

Zoom In Zoom Out Reset image size

After the transistors start switching, the L–C circuit begins oscillation with increasing amplitude until reaching steady state. The envelope of this transient can be fitted with an exponential rise and decay at the beginning and end of the stimulation pulse, with time constant inversely proportional to the frequency. Ch 1's time constant was approximately 0.2 ms, requiring about 1 ms to reach steady state or for the oscillation to stop (figure 3(A)); for Ch 2 and Ch 3, the transient's durations were 10- and 100-times shorter, respectively, and therefore negligible compared to the minimum pulse duration anticipated (10 ms). Within each channel, the pulse can be turned on and off with sub-millisecond speed. Between channels, the switching was slower for the contactors to close or open. Also, for safety reasons, a short delay was inserted in the control program when switching between different channels to avoid the high current from one channel and voltage from one capacitor bank carrying over to another one. The switching between channels can be achieved as fast as 1 ms (figure 3(B)).

Zoom In Zoom Out Reset image size

Ch 1 and Ch 2 are thermally limited by temperature rise in the Si transistors, the series AC capacitors, the cable and the coil, and may run single pulses up to 10 s and 30 s duration for the presented current levels, respectively. Ch 3 can achieve continuous operation. To test the stability of the system, a pulse train cycling through the three channels was output with equal pulse duration and interpulse intervals at two different duty cycles. The system was able to continuously operate for two hours, with the temperature reaching a steady cycle around 20 min. All components operated well within their specifications and, besides the cable, had moderate temperature (figure S5). As a potential bottleneck of thermal performance, the cable could be made thicker to reduce power losses and heating.

The heating of the three types of nanoparticles was measured over 3 s or 15 s magnetic field stimulation in all three channels, and their specific absorption rates are calculated (figure 4 and table S3). When stimulation time is limited to 3 s for Ch 1 and 15 s for Ch 2 and 3, the maximum temperature change showed a selectivity of ∼9× for 15 nm Cobalt doped nanoparticles in Ch 1, ∼2.3× for 19 nm iron-oxide nanoparticles in Ch 2, and ∼3.7× for 10 nm Mn-doped nanoparticles in Ch 3, thus indicating three distinct channels available for magnetothermal heating. By adjusting the concentration of the nanoparticles and the corresponding stimulation time duration, three samples within the same magnetic field could be separately activated (figure 5).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size