Redox‐Addressable Single‐Molecule Junctions Incorporating a Persistent Organic Radical

Abstract Integrating radical (open‐shell) species into non‐cryogenic nanodevices is key to unlocking the potential of molecular electronics. While many efforts have been devoted to this issue, in the absence of a chemical/electrochemical potential the open‐shell character is generally lost in contact with the metallic electrodes. Herein, single‐molecule devices incorporating a 6‐oxo‐verdazyl persistent radical have been fabricated using break‐junction techniques. The open‐shell character is retained at room temperature, and electrochemical gating permits in situ reduction to a closed‐shell anionic state in a single‐molecule transistor configuration. Furthermore, electronically driven rectification arises from bias‐dependent alignment of the open‐shell resonances. The integration of radical character, transistor‐like switching, and rectification in a single molecular component paves the way to further studies of the electronic, magnetic, and thermoelectric properties of open‐shell species.


Additional Experimental Details
Compounds 1 and 2 were prepared as described elsewhere. [1] Cyclic voltammetry measurements plotted in Figure 3 of the manuscript were conducted in a standard three-electrode cell, with Pt disc working electrode, Pt wire counter and Pt wire pseudo-reference electrodes, from solutions in 0.1 M NBu4PF6 / CH2Cl2, with data collected from an EmStat3+ potentiostat. The ferrocene/ ferrocenium (E1/2 = 0 V) and decamethyl ferrocene/ decamethyl ferrocenium couples (E1/2 = -0.55 V versus ferrocene / ferrocenium) were used as internal references for potential measurements. [2] 2. EPR Methods and Data Liquid solution EPR samples: A solution of 1 was prepared in 100 µM concentration in dichloromethane (CH2Cl2), degassed by bubbling under Argon gas for 15 minutes. The sample was measured in a Q-band capillary (0.3 mm ID), placed inside an X-band quartz tube to minimize microwave dampening by CHCl3.

Frozen solution EPR samples:
A solution of 1 was prepared in 100 µM concentration 50/50 mix of CHCl3/toluene. The sample was measured in a 3.8 mm OD X-band EPR tube at 120 K.
Gold/substrate EPR samples: Sample of 1 deposited onto Au coated glass substrate were also prepared for cw-EPR measurements. Substrates were incubated in 1 mM solution of 1 in chloroform for 48 hours, followed by rinsing and drying under flowed nitrogen gas. The gold-plated glass substrate was then inserted into a 3.8 mm OD X-band EPR tube.
Continuous Wave cw-EPR: measurements were performed at room temperature on a Bruker E500 spectrometer equipped with an ER4122 SHQ resonator. Spectra were recorded with a field modulation amplitude of 0.5 G and a power of 0.47 mW for the solution phase measurements, and a modulation amplitude of up 5 G and 4.7 mW for the gold/substrate measurements.
EPR simulations: Spectra were simulated using the EasySpin package [3] in MATLAB using the isotropic and fast-motion cw-EPR program garlic for the solution spectrum, and pepper for the solid state spectrum. Simulations were optimized using a Nelder-Mead minimization algorithm.
EPR data:The solution EPR spectrum of 1 in dichloromethane at room temperature is shown in Figure S1A. Its lineshape is typical of 6-oxo-verdazyls and its derivatives. [4,5] The nine-line structure comes about from the coupling of the unpaired electron spin to the four 14 N nuclei (I = 1) of the central ring. [4] Owing to the symmetry of the molecule, the four nitrogens form two equivalent sets (Table S1). As shown in a previous publication by our group, the introduced phenyl rings at the 1-and 5-positions of the 6-oxo-verdazyl core has little effect on its electronic structure. [6] The frozen solution (solid state) EPR spectrum of 1 was also measured ( Figure S1B). Owing to inhomogeneous line broadening and the molecules now no longer tumbling in solution, the EPR spectrum is characteristically less resolved, although some 14 N hyperfine structure is retained. Its g value (central crossing point) is slightly higher than in solution (2.005 vs. 2.0039). to gold plated glass baseline (blue). The difference between the two spectra is shown in red.
We also measured the EPR spectrum of 1 deposited on a gold surface ( Figure S1B). A radical signal is observed, which appears at the same g value of g~2.005 as the radical measured in frozen solution. The bare Au-coated substrate shows no large EPR signal in this spectral region, confirming the recorded signal arises from successful chemisorption and retention of 1 on the gold surface. The EPR spectrum of 1 on Au is of a similar width to the frozen solution spectrum, but does not clearly resolve any 14 N hyperfine structure, suggesting the unpaired electron density across 1 may be more delocalized when attached to the gold surface. No significant orientation dependence of the gold plates was observed. At low temperature (~100 K) the radical signal was reduced significantly, supporting the assignment to a slow relaxing organic radical species.

Additional Details:
A basis set that describes the radical spin manifold can be built from the product of the eigenstates of the interacting electron (S = 1/2) and nuclear ( 14 N, I = 1) spins: of 1 and mi the values -1, 0 and +1. The spin Hamiltonian that describes the spin manifold is: Eq. S2 It contains (i) an electronic Zeeman term describing the unpaired electron's interaction with the applied magnetic field (g), (iv) a nuclear Zeeman term for each 14 N nucleus and the applied magnetic field, and (iii) an electron-nuclear hyperfine term (ai) for each 14 N nucleus describing the magnetic interaction between the unpaired electron and each nucleus. Note that the nuclear quadruople term does not need to be considered when simulation the EPR spectrum.
Owing to the symmetry of the oxoverdazyl framework, the four 14 N hyperfine couplings represent two equivalent sets: two larger couplings a1 and a2; and two smaller couplings a3 and a4. In Table S1, a1 and a2 are described by the label a(N2, 4) and a3 and a4 are described by the label a(N1,5) which makes use of the crystallographic labelling of the four nitrogen sites.

Single-Molecule Conductance Measurements
A modified Keysight 5500 STM was employed for the fabrication and characterisation of single-molecule junctions.
STMBJ and EC-STMBJ (electrochemical control) experiments were performed using a custom 4-channel current amplifier based on the design by Meszaros et al. [7] The Keysight N9610A electronics controls the tip position and substrate bias, while the electrochemical potentials are applied by a bipotentiostat integrated in the STM controller. The 4-channel output of the preamplifier is recorded at 10 kSa/s by a National Instruments NI9215 USB DAQ with bespoke Python software. insulating wax (Apiezon W). A coiled platinum wire (Goodfellow 99.99+%) was used as counterelectrode and an electrochemically chloridised Ag wire (Goodfellow 99.999+%) was used as reference electrode. [8] In all experiments the tip is held at ground.
For the determination of molecular conductance using regular STMBJ methods, all acquired data was used without further selection. For the measurements of the − characteristics, we used a method described in details elsewhere, [9] with data collected from junction formed with bare-wire tips to avoid two-electrode electrochemistry In brief, data is acquired using a staircase ramp applied to the piezo transducer, with abrupt stretches of 1.1 nm followed by a 100 ms "hold" portion. During the hold, the bias is held at a constant value of 200 mV for 25 ms, and then ramped between 2 and -2 V at a rate of 80 V/s. After the ramp, the bias is held again at a constant value of 200 mV until the end of the hold. An automated algorithm was then used to analyse the data. First, the algorithm sliced the traces between abrupt stretches, by calculating the second derivative of the piezo signal and cutting when its value went above a threshold. After that, the average conductance of the junction during the fixed bias sections is calculated, and the algorithm only selects traces where this value falls within one standard deviation from the most probable molecular conductance, as determined by regular break-junction measurements. This process filters out data where no junctions was formed, or where the molecular junction did not survive the whole bias modulation process. On average, 30-40% of traces are retained by our algorithm. The resulting slices, now only relative to stable junctions are used without further processing and compiled into 2d heatmaps as shown in the manuscript and later in this document.

Additional STM-BJ Data
In addition to the data presented in the manuscript, additional data and details are provided here. 2D maps and full STMBJ histograms are presented below, for experiments performed in mesitylene. Single-molecule electrochemical break-junction experiments were also performed over a wider potential range than shown in the main manuscript, exploring an electrochemical window 1.8 V wide. Data at potentials > 0.5 V vs Fc/Fc + showed greatly increased instrumental noise and no apparent junction formation. From the cyclic voltammetry shown in the main paper we expect 1 to be oxidised at these potentials, but the technical difficulties detailed earlier prevented us from characterising the charge transport properties in the +1 state. It is indeed possible that 1 in its cationic state is not soluble in the ionic liquid we employed, and therefore precipitates out of solution and/or deposits on the electrode surface.
A wide-range conductance vs potential map is presented in Figure S6. It should be noted here that no data could be acquired at potentials > 0.  Figure S6: Single-molecule conductance data for 1 across the full electrochemical window explored. 0.2 V bias, 1 mM in 1-butyl-methylimidazolium triflate, Pt counterelectrode, Ag/AgCl reference electrode.
In addition to the full electrochemical map, we present here data collected after cycling 1 between the radical and the anionic state. Upon return to a small negative potential, the high conductance signal could be recovered.

Additional Electrochemical Data
The electrochemical behaviour of 1 is shown in the main paper on a Pt electrode, in CH2Cl2 with tetrabutylammonium hexafluorophosphate as support electrolyte. We also recorded CVs of 1 in the environment used for the STMBJ measurements, using an Au working electrode, a Pt counter-electrode, and a Pt pseudo-reference electrode in 1-butyl-3methylimidazolium triflate. As can be observed in Figure S10, the reduction of 1 retains good chemical reversibility and quasi-reversible electrochemical behaviour. 1 can be cycled between the neutral radical and the anionic state as shown by the wave in the CV centred at around -900 mV (vs. Fc/Fc + ), with a peak-to-peak separation of 140 mV at a scan speed of 70 mV/s. The oxidation of 1 to the cationic state shows a peak at about 0 V at lower sweep rates, which is, however, completely irreversible. As discussed in the main text, we attribute this to the insolubility of the cationic species in 1-butyl-3methylimidazolium triflate, leading to precipitation and preventing its reduction as the voltage is made less positive.

Computational Methods
Characterisation of gas phase molecules: The optimized geometry of 1 and 2, their molecular orbitals and spin density of 1 are calculated density functional theory. We perform these calculations with two different codes. SIESTA (using parameters described below) and Gaussian v16 (with B3LYP hybrid functional, qzvp basis set and tight convergence criteria) and obtained similar results.
Molecules between gold electrodes: The optimized geometry and ground state Hamiltonian and overlap matrix elements of each structure studied in this paper was self-consistently obtained using the SIESTA implementation [10] of density functional theory (DFT). SIESTA employs norm-conserving pseudo-potentials to account for the core electrons and linear combinations of atomic orbitals to construct the valence states. The local density approximation (LDA) of the exchange and correlation functional is used with the CA parameterization a double-ζ polarized (DZP) basis set, a realspace grid defined with an equivalent energy cut-off of 150 Ry. The geometry optimization for each structure is performed to the forces smaller than 20 meV/Å.

Quantum transport calculations:
The mean-field Hamiltonian obtained from the converged DFT calculation was combined with GOLLUM [11,12] implementation of the non-equilibrium Green's function method [12] to calculate the phase- We note that the actual electrode structure is unknown in the experiment, therefore it is not clear that how much of potential drops happens across the junction, at the electrode and its surface and at the interface between molecule and electrodes. Therefore, we can only compare qualitatively our bias dependent calculations with the experimental results.
That is why we choose a ±1 V bias window and demonstrate that the resonances due to frontier orbitals are shifted by the