A Scalable Approach to Modeling on Accelerated Neuromorphic Hardware.

Neuromorphic systems open up opportunities to enlarge the explorative space for computational research. However, it is often challenging to unite efficiency and usability. This work presents the software aspects of this endeavor for the BrainScaleS-2 system, a hybrid accelerated neuromorphic hardware architecture based on physical modeling. We introduce key aspects of the BrainScaleS-2 Operating System: experiment workflow, API layering, software design, and platform operation. We present use cases to discuss and derive requirements for the software and showcase the implementation. The focus lies on novel system and software features such as multi-compartmental neurons, fast re-configuration for hardware-in-the-loop training, applications for the embedded processors, the non-spiking operation mode, interactive platform access, and sustainable hardware/software co-development. Finally, we discuss further developments in terms of hardware scale-up, system usability, and efficiency.

The BrainScaleS-2 Accelerated Neuromorphic System With Hybrid Plasticity.

Since the beginning of information processing by electronic components, the nervous system has served as a metaphor for the organization of computational primitives. Brain-inspired computing today encompasses a class of approaches ranging from using novel nano-devices for computation to research into large-scale neuromorphic architectures, such as TrueNorth, SpiNNaker, BrainScaleS, Tianjic, and Loihi. While implementation details differ, spiking neural networks-sometimes referred to as the third generation of neural networks-are the common abstraction used to model computation with such systems. Here we describe the second generation of the BrainScaleS neuromorphic architecture, emphasizing applications enabled by this architecture. It combines a custom analog accelerator core supporting the accelerated physical emulation of bio-inspired spiking neural network primitives with a tightly coupled digital processor and a digital event-routing network.

Surrogate gradients for analog neuromorphic computing.

To rapidly process temporal information at a low metabolic cost, biological neurons integrate inputs as an analog sum, but communicate with spikes, binary events in time. Analog neuromorphic hardware uses the same principles to emulate spiking neural networks with exceptional energy efficiency. However, instantiating high-performing spiking networks on such hardware remains a significant challenge due to device mismatch and the lack of efficient training algorithms. Surrogate gradient learning has emerged as a promising training strategy for spiking networks, but its applicability for analog neuromorphic systems has not been demonstrated. Here, we demonstrate surrogate gradient learning on the BrainScaleS-2 analog neuromorphic system using an in-the-loop approach. We show that learning self-corrects for device mismatch, resulting in competitive spiking network performance on both vision and speech benchmarks. Our networks display sparse spiking activity with, on average, less than one spike per hidden neuron and input, perform inference at rates of up to 85,000 frames per second, and consume less than 200 mW. In summary, our work sets several benchmarks for low-energy spiking network processing on analog neuromorphic hardware and paves the way for future on-chip learning algorithms.

Washout and Awakening Times after Inhaled Sedation of Critically Ill Patients: Desflurane Versus Isoflurane.

In recent years, inhaled sedation has been increasingly used in the intensive care unit (ICU). The aim of this prospective, controlled trial was to compare washout and awakening times after long term sedation with desflurane and isoflurane both administered with the Mirus™ system (TIM GmbH, Koblenz, Germany). Twenty-one consecutive critically ill patients were alternately allocated to the two study groups, obtaining inhaled sedation with either desflurane or isoflurane. After 24 h study sedation, anesthetic washout curves were recorded, and a standardized wake-up test was performed. The primary outcome measure was the time required to decrease the endtidal concentration to 50% (T50%). Secondary outcome measures were T80% and awakening times (all extremities moved, RASS -2). Decrement times (min) (desflurane versus isoflurane, median (1st quartile-3rd quartile)) (T50%: 0.3 (0.3-0.4) vs. 1.3 (0.4-2.3), log-rank test P = 0.002; P80%: 2.5 (2-5.9) vs. 12.1 (5.1-20.2), P = 0.022) and awakening times (to RASS -2: 7.5 (5.5-8.8) vs. 41.0 (24.5-43.0), P = 0.007; all extremities moved: 5.0 (4.0-8.5) vs. 13.0 (8.0-41.25), P = 0.037) were significantly shorter after desflurane compared to isoflurane. The use of desflurane with the Mirus™ system significantly shortens the washout times and leads to faster awakening after sedation of critically ill patients.

Photochemical reactions on aerosols at West Antarctica: A molecular case-study of nitrate formation among sea salt aerosols.

Environmental implications of climate change are complex and exhibit regional variations both within and between the polar regions. The increase of solar UV radiation flux over Antarctica due to stratospheric ozone depletion creates the optimal conditions for photochemical reactions on the snow. Modeling, laboratory, and indirect field studies suggest that snowpack process release gases to the atmosphere that can react on sea salt particles in remote regions such as Antarctica, modifying aerosol composition and physical properties of aerosols. Here, we present evidence of photochemical processing in West Antarctica aerosols using microscopic and chemical speciation of individual atmospheric particles. Individual aerosol particles collected at the Brazilian module Criosfera 1 were analyzed by scanning transmission X-ray microscopy with near edge X-ray absorption fine structure spectroscopy (STXM/NEXAFS) combined with computer-controlled scanning electron microscopy (CCSEM) with energy-dispersive X-ray (EDX) microanalysis. The displacement of chlorine relative to sodium was observed over most of the sea salt particles. Particles with a chemical composition consistent with NaCl-NO3 contributed up to 30% of atmospheric particles investigated. Overall, this study provides evidence that the snowpack and particulate nitrate photolysis should be considered in dynamic partition equilibrium in the troposphere. These findings may assist in reducing modeling uncertainties and present new insights into the aerosol chemical composition in the polar environment.

Revealing Nanoscale Chemical Heterogeneities in Polycrystalline Mo-BiVO4 Thin Films.

The activity of polycrystalline thin film photoelectrodes is impacted by local variations of the material properties due to the exposure of different crystal facets and the presence of grain/domain boundaries. Here a multi-modal approach is applied to correlate nanoscale heterogeneities in chemical composition and electronic structure with nanoscale morphology in polycrystalline Mo-BiVO4 . By using scanning transmission X-ray microscopy, the characteristic structure of polycrystalline film is used to disentangle the different X-ray absorption spectra corresponding to grain centers and grain boundaries. Comparing both spectra reveals phase segregation of V2 O5 at grain boundaries of Mo-BiVO4 thin films, which is further supported by X-ray photoelectron spectroscopy and many-body density functional theory calculations. Theoretical calculations also enable to predict the X-ray absorption spectral fingerprint of polarons in Mo-BiVO4 . After photo-electrochemical operation, the degraded Mo-BiVO4 films show similar grain center and grain boundary spectra indicating V2 O5 dissolution in the course of the reaction. Overall, these findings provide valuable insights into the degradation mechanism and the impact of material heterogeneities on the material performance and stability of polycrystalline photoelectrodes.

Fungal spores as a source of sodium salt particles in the Amazon basin.

In the Amazon basin, particles containing mixed sodium salts are routinely observed and are attributed to marine aerosols transported from the Atlantic Ocean. Using chemical imaging analysis, we show that, during the wet season, fungal spores emitted by the forest biosphere contribute at least 30% (by number) to sodium salt particles in the central Amazon basin. Hydration experiments indicate that sodium content in fungal spores governs their growth factors. Modeling results suggest that fungal spores account for ~69% (31-95%) of the total sodium mass during the wet season and that their fractional contribution increases during nighttime. Contrary to common assumptions that sodium-containing aerosols originate primarily from marine sources, our results suggest that locally-emitted fungal spores contribute substantially to the number and mass of coarse particles containing sodium. Hence, their role in cloud formation and contribution to salt cycles and the terrestrial ecosystem in the Amazon basin warrant further consideration.

Heating-Induced Transformations of Atmospheric Particles: Environmental Transmission Electron Microscopy Study.

Environmental transmission electron microscopy was employed to probe transformations in the size, morphology, and composition of individual atmospheric particles as a function of temperature. Two different heating devices were used and calibrated in this work: a furnace heater and a Micro Electro Mechanical System heater. The temperature calibration used sublimation temperatures of NaCl, glucose, and ammonium sulfate particles, and the melting temperature of tin. Volatilization of Suwanee River Fulvic Acid was further used to validate the calibration up to 800 °C. The calibrated furnace holder was used to examine both laboratory-generated secondary organic aerosol particles and field-collected atmospheric particles. Chemical analysis by scanning transmission X-ray microscopy and near-edge fine-structure spectroscopy of the organic particles at different heating steps showed that above 300 °C particle volatilization was accompanied by charring. These methods were then applied to ambient particles collected in the central Amazon region. Distinct categories of particles differed in their volatilization response to heating. Spherical, more-viscous particles lost less volume during heating than particles that spread on the imaging substrate during impaction, due to either being liquid upon impaction or lower viscosity. This methodology illustrates a new analytical approach to accurately measure the volume fraction remaining for individually tracked atmospheric particles at elevated temperatures.

Rupturing of Biological Spores As a Source of Secondary Particles in Amazonia.

Airborne biological particles, such as fungal spores and pollen, are ubiquitous in the Earth's atmosphere and may influence the atmospheric environment and climate, impacting air quality, cloud formation, and the Earth's radiation budget. The atmospheric transformations of airborne biological spores at elevated relative humidity remain poorly understood and their climatic role is uncertain. Using an environmental scanning electron microscope (ESEM), we observed rupturing of Amazonian fungal spores and subsequent release of submicrometer size fragments after exposure to high humidity. We find that fungal fragments contain elements of inorganic salts (e.g., Na and Cl). They are hygroscopic in nature with a growth factor up to 2.3 at 96% relative humidity, thus they may potentially influence cloud formation. Due to their hygroscopic growth, light scattering cross sections of the fragments are enhanced by up to a factor of 10. Furthermore, rupturing of fungal spores at high humidity may explain the bursting events of new particle formation in Amazonia.