Two-dimensional fully ferroelectric-gated hybrid computing-in-memory hardware for high-precision and energy-efficient dynamic tracking.

Computing in memory (CIM) breaks the conventional von Neumann bottleneck through in situ processing. Monolithic integration of digital and analog CIM hardware, ensuring both high precision and energy efficiency, provides a sustainable paradigm for increasingly sophisticated artificial intelligence (AI) applications but remains challenging. Here, we propose a complementary metal-oxide semiconductor-compatible ferroelectric hybrid CIM platform that consists of Boolean logic and triggers for digital processing and multistage cell arrays for analog computation. The basic ferroelectric-gated units are assembled with solution-processable two-dimensional (2D) molybdenum disulfide atomic-thin channels at a wafer-scale yield of 96.36%, delivering high on/off ratios (>107), high endurance (>1012), long retention time (>10 years), and ultralow cycle-to-cycle/device-to-device variations (~0.3%/~0.5%). Last, we customize a highly compact 2D hybrid CIM system for dynamic tracking, achieving a high accuracy of 99.8% and a 263-fold improvement in power efficiency compared to graphics processing units. These results demonstrate the potential of 2D fully ferroelectric-gated hybrid hardware for developing versatile CIM blocks for AI tasks.

Microbial perspective on the skin-gut axis and atopic dermatitis.

Atopic dermatitis (AD) is a relapsing inflammatory skin condition that has become a global health issue with complex etiology and mounting prevalence. The association of AD with skin and gut microbiota has been revealed by virtue of the continuous development of sequencing technology and genomics analysis. Also, the gut-brain-skin axis and its mutual crosstalk mechanisms have been gradually verified. Accordingly, the microbiota-skin-gut axis also plays an important role in allergic skin inflammation. Herein, we reviewed the relationship between the microbiota-skin-gut axis and AD, explored the underlying signaling molecules and potential pathways, and focused on the potential mechanisms of probiotics, antimicrobial peptides (AMPs), coagulase-negative staphylococci transplantation, fecal microbiota transplantation, AMPs, and addition of essential fatty acids in alleviating AD, with the aim to provide a new perspective for targeting microbiota in the treatment of allergic skin inflammation.

Monolithic three-dimensional integration of RRAM-based hybrid memory architecture for one-shot learning.

In this work, we report the monolithic three-dimensional integration (M3D) of hybrid memory architecture based on resistive random-access memory (RRAM), named M3D-LIME. The chip featured three key functional layers: the first was Si complementary metal-oxide-semiconductor (CMOS) for control logic; the second was computing-in-memory (CIM) layer with HfAlOx-based analog RRAM array to implement neural networks for feature extractions; the third was on-chip buffer and ternary content-addressable memory (TCAM) array for template storing and matching, based on Ta2O5-based binary RRAM and carbon nanotube field-effect transistor (CNTFET). Extensive structural analysis along with array-level electrical measurements and functional demonstrations on the CIM and TCAM arrays was performed. The M3D-LIME chip was further used to implement one-shot learning, where ~96% accuracy was achieved on the Omniglot dataset while exhibiting 18.3× higher energy efficiency than graphics processing unit (GPU). This work demonstrates the tremendous potential of M3D-LIME with RRAM-based hybrid memory architecture for future data-centric applications.

In-Memory Realization of Eligibility Traces Based on Conductance Drift of Phase Change Memory for Energy-Efficient Reinforcement Learning.

Reinforcement learning (RL) has shown outstanding performance in handling complex tasks in recent years. Eligibility trace (ET), a fundamental and important mechanism in reinforcement learning, records critical states with attenuation and guides the update of policy, which plays a crucial role in accelerating the convergence of RL training. However, ET implementation on conventional digital computing hardware is energy hungry and restricted by the memory wall due to massive calculation of exponential decay functions. Here, in-memory realization of ET for energy-efficient reinforcement learning with outstanding performance in discrete- and continuous-state RL tasks is demonstrated. For the first time, the inherent conductance drift of phase change memory is exploited as physical decay function to realize in-memory eligibility trace, demonstrating excellent performance during RL training in various tasks. The spontaneous in-memory decay computing and storage of policy in the same phase change memory give rise to significantly enhanced energy efficiency compared with traditional graphics processing unit platforms. This work therefore provides a holistic energy and hardware efficient method for both training and inference of reinforcement learning.