Search for: All records

Abstract Ultra-thin films of low damping ferromagnetic insulators with perpendicular magnetic anisotropy have been identified as critical to advancing spin-based electronics by significantly reducing the threshold for current-induced magnetization switching while enabling new types of hybrid structures or devices. Here, we have developed a new class of ultra-thin spinel structure Li_0.5Al_1.0Fe_1.5O₄(LAFO) films on MgGa₂O₄(MGO) substrates with: 1) perpendicular magnetic anisotropy; 2) low magnetic damping and 3) the absence of degraded or magnetic dead layers. These films have been integrated with epitaxial Pt spin source layers to demonstrate record low magnetization switching currents and high spin-orbit torque efficiencies. These LAFO films on MGO thus combine all of the desirable properties of ferromagnetic insulators with perpendicular magnetic anisotropy, opening new possibilities for spin based electronics. 

Neural networks tend to achieve better accuracy with training if they are larger — even if the resulting models are overparameterized. Nevertheless, carefully removing such excess of parameters before, during, or after training may also produce models with similar or even improved accuracy. In many cases, that can be curiously achieved by heuristics as simple as removing a percentage of the weights with the smallest absolute value — even though absolute value is not a perfect proxy for weight relevance. With the premise that obtaining significantly better performance from pruning depends on accounting for the combined effect of removing multiple weights, we revisit one of the classic approaches for impact-based pruning: the Optimal Brain Surgeon (OBS). We propose a tractable heuristic for solving the combinatorial extension of OBS, in which we select weights for simultaneous removal, and we combine it with a single-pass systematic update of unpruned weights. Our selection method outperforms other methods for high sparsity, and the single-pass weight update is also advantageous if applied after those methods. 

We can compress a rectifier network while exactly preserving its underlying functionality with respect to a given input domain if some of its neurons are stable. However, current approaches to determine the stability of neurons with Rectified Linear Unit (ReLU) activations require solving or finding a good approximation to multiple discrete optimization problems. In this work, we introduce an algorithm based on solving a single optimization problem to identify all stable neurons. Our approach is on median 183 times faster than the state-of-art method on CIFAR-10, which allows us to explore exact compression on deeper (5 x 100) and wider (2 x 800) networks within minutes. For classifiers trained under an amount of L1 regularization that does not worsen accuracy, we can remove up to 56% of the connections on the CIFAR-10 dataset. The code is available at the following link, https://github.com/yuxwind/ExactCompression 

Abstract Organic semiconductors with high-spin ground states are fascinating because they could enable fundamental understanding on the spin-related phenomenon in light element and provide opportunities for organic magnetic and quantum materials. Although high-spin ground states have been observed in some quinoidal type small molecules or doped organic semiconductors, semiconducting polymers with high-spin at their neutral ground state are rarely reported. Here we report three high-mobility semiconducting polymers with different spin ground states. We show that polymer building blocks with small singlet-triplet energy gap (Δ E S-T ) could enable small Δ E S-T gap and increase the diradical character in copolymers. We demonstrate that the electronic structure, spin density, and solid-state interchain interactions in the high-spin polymers are crucial for their ground states. Polymers with a triplet ground state ( S  = 1) could exhibit doublet ( S  = 1/2) behavior due to different spin distributions and solid-state interchain spin-spin interactions. Besides, these polymers showed outstanding charge transport properties with high hole/electron mobilities and can be both n- and p-doped with superior conductivities. Our results demonstrate a rational approach to obtain high-mobility semiconducting polymers with different spin ground states. 

Abstract New stimulus‐responsive scaffolds are of interest as constituents of hierarchical supramolecular ensembles. 1,3,5–2,4,6‐Functionalized, facially segregated benzene moieties have a time‐honored role as building blocks for host molecules. However, their user as switchable motifs in the construction of multi‐component supramolecular structures remains poorly explored. Here, we report a molecular cage 1, which consists of a bent anthracene dimer3paired with 1,3,5‐tris(aminomethyl)‐2,4,6‐triethylbenzene2. As the result of the pH‐inducedababab↔bababaisomerization of the constituent‐functionalized benzene units derived from2, this cage can reversibly convert between an open state and a closed form, both in solution and in the solid state. Cage 1was used to create stimuli‐responsive hierarchical superstructures, namely Russian doll‐like complexes with [K⊂18‐crown‐6⊂1]⁺and [K⊂cryptand‐222⊂1]⁺. The reversible assembly and disassembly of these superstructures could be induced by switching cage 1from its open to closed form. The present study thus provides an unusual example where pH‐triggered conformation motion within a cage‐like scaffold is used to control the formation and disassociation of hierarchical ensembles.