Foundation models (FMs) are achieving remarkable successes to realize complex downstream tasks in domains including natural language and visions. In this paper, we propose building an FM for material science, which is trained with massive data across a wide variety of material domains and data modalities. Nowadays machine learning models play key roles in material discovery, particularly for property prediction and structure generation. However, those models have been independently developed to address only specific tasks without sharing more global knowledge. Development of an FM for material science will enable overarching modeling across material domains and data modalities by sharing their feature representations. We discuss fundamental challenges and required technologies to build an FM from the aspects of data preparation, model development, and downstream tasks.

The discovery of new materials has been the essential force which brings a discontinuous improvement to industrial products' performance. However, the extra-vast combinatorial design space of material structures exceeds human experts' capability to explore all, thereby hampering material development. In this paper, we present a material industry-oriented web platform of an AI-driven molecular inverse-design system, which automatically designs brand new molecular structures rapidly and diversely. Different from existing inverse-design solutions, in this system, the combination of substructure-based feature encoding and molecular graph generation algorithms allows a user to gain high-speed, interpretable, and customizable design process. Also, a hierarchical data structure and user-oriented UI provide a flexible and intuitive workflow. The system is deployed on IBM's and our client's cloud servers and has been used by 5 partner companies. To illustrate actual industrial use cases, we exhibit inverse-design of sugar and dye molecules, that were carried out by experimental chemists in those client companies. Compared to a general human chemist's standard performance, the molecular design speed was accelerated more than 10 times, and greatly increased variety was observed in the inverse-designed molecules without loss of chemical realism.

The development of hardware neural networks, including neuromorphic hardware, has been accelerated over the past few years. However, it is challenging to operate very large-scale neural networks with low-power hardware devices, partly due to signal transmissions through a massive number of interconnections. Our aim is to deal with the issue of communication cost from an algorithmic viewpoint and study learning algorithms for energy-efficient information processing. Here, we consider two approaches to finding spatially arranged sparse recurrent neural networks with the high cost-performance ratio for associative memory. In the first approach following classical methods, we focus on sparse modular network structures inspired by biological brain networks and examine their storage capacity under an iterative learning rule. We show that incorporating long-range intermodule connections into purely modular networks can enhance the cost-performance ratio. In the second approach, we formulate for the first time an optimization problem where the network sparsity is maximized under the constraints imposed by a pattern embedding condition. We show that there is a tradeoff between the interconnection cost and the computational performance in the optimized networks. We demonstrate that the optimized networks can achieve a better cost-performance ratio compared with those considered in the first approach. We show the effectiveness of the optimization approach mainly using binary patterns and apply it also to gray-scale image restoration. Our results suggest that the presented approaches are useful in seeking more sparse and less costly connectivity of neural networks for the enhancement of energy efficiency in hardware neural networks.

Reservoir computing is a computational framework suited for temporal/sequential data processing. It is derived from several recurrent neural network models, including echo state networks and liquid state machines. A reservoir computing system consists of a reservoir for mapping inputs into a high-dimensional space and a readout for pattern analysis from the high-dimensional states in the reservoir. The reservoir is fixed and only the readout is trained with a simple method such as linear regression and classification. Thus, the major advantage of reservoir computing compared to other recurrent neural networks is fast learning, resulting in low training cost. Another advantage is that the reservoir without adaptive updating is amenable to hardware implementation using a variety of physical systems, substrates, and devices. In fact, such physical reservoir computing has attracted increasing attention in diverse fields of research. The purpose of this review is to provide an overview of recent advances in physical reservoir computing by classifying them according to the type of the reservoir. We discuss the current issues and perspectives related to physical reservoir computing, in order to further expand its practical applications and develop next-generation machine learning systems.

A new architecture based on multi-mode optical interconnect technology is proposed and described for the realization of a compact time delay opto-electronic reservoir computer. VCSEL and photodiode chip arrays are flip-chip mounted on a polymer waveguide structure containing micro-mirrors. A node is made of a photodiode and a laser connected back-to-back, with an amplified signal. The specifications for this amplifier are investigated and it is found that the cutoff frequency can be lower than the incoming sample rate for signal recovery and chaotic time series prediction tasks. An optical waveguide delay line sample is also presented and characterized experimentally. Results are promising for the realization of a compact platform for reservoir computers that could easily be scaled up to several physical nodes.

In this paper we report on a wireless Ethernet link over 60GHz mmWave, which is a full-duplex wireless link in the same frequency channel. To provide a dense wireless network, beam forming with a phased-array transceiver is a key technology for frequency reuse. We introduced a particle swarm optimization algorithm for beam forming with effective interference cancellation in each field. This method works independent of the transceiver arrangement and is free from device-specific variability and non-linearity. We prototyped the wireless Ethernet link system and achieved a BER less than 1.0E-11 at a 3m distance. We were able to prove its feasibility at system-level application.

This study explores efficient structures of artificial neural networks for associative memories. Motivated by the real brain structure and the demand of energy efficiency in hardware implementation, we consider neural networks with sparse modular structures. Numerical experiments are performed to clarify how the storage capacity of associative memory depends on regularity and randomness of the network structures. We first show that a fully regularized network, suited for design of hardware, has poor recall performance and a fully random network, undesired for hardware implementation, yields excellent recall performance. For seeking a network structure with good performance and high implementability, we consider four different modular networks constructed based on different combinations of regularity and randomness. From the results of associative memory tests for these networks, we find that the combination of random intramodule connections and regular intermodule connections works better than the other cases. Our results suggest that the parallel usage of regularity and randomness in network structures could be beneficial for developing energy-efficient neural networks.

This paper describes a millimeter wave wireless system using the 60-GHz frequency band, focusing on the data recovery algorithm and implementation to compensate for both the phase noise and the DC offset. Our robust baseband system supports data transfers at 7.04 Gbps using 16-QAM single-carrier with effective digital compensation, an improved adaptive equalizer, and a post-phase rotator and post-shifter. We use a low-overhead frame format without training symbols in the payload and RS(255, 239) as the error correction code so that we can retain spectral efficiency with an effective data rate around 6.6 Gbps with a bit error rate of 4.0E-5 at 3 m distance. Our results show how millimeter wave technology can provide higher data rates and higher aggregated channel capacities for future dense networks.

This paper presents a high-speed multimedia content-downloading system using 60GHz millimeter wave (mmWave) technology. For a cost-effective solution, the system makes use of conventional 802.11 for the interactive controls between an access point and smart mobile devices, handling such services as access point discovery, authentication, mmWave link establishment, beam steering, and lost packet recovery. The high-speed one-way mmWave link is dedicated to the transfer of content data. The smart mobile device only has an mmWave receiver so that it can keep its power dissipation and hardware costs low. The content data is downloaded from an access point to such a smart mobile device via an external storage component of the mobile device. To use the high bandwidth of the mmWave, the mmWave receiver writes directly to the external storage as a wireless attached memory, which is accessed by the smart mobile device via its SD interface. We built a proof-of-concept prototype with a smartphone receiving data from an mmWave system, and evaluated the performance.

This paper deals with mm Wave MIMO link design strategy for wireless data center applications where the MIMO degrees of freedom is taken into account in multi-node packet networking environments. The problem is treated differently from the coordinated multiuser MIMO situation and the link design is optimized independently in each node with meeting control and data plane requirements for contention-based packet switching. In particular, we propose using an interference-aligned out-of-band control plane to improve the unidirectional bonded in-band data plane collision-related performance degradation with a limited number of antenna elements per node. We also present a high-level implementation plan.

This paper presents a cost-effective high-speed multimedia-content download system using a hybrid wireless link consisting of a 60-GHz (mmWave) link and a conventional wireless link, such as IEEE 802.11. The capability gap between the components of existing systems and a mmWave link caps system-level performance and degrades resource utilization. We developed an FPGA-based prototype system to investigate system-level performance for well balanced systems. Our performance evaluation involved changing the PHY parameters, data acknowledgement (ACK) methods and link quality. Under the point-to-point network, our results confirmed that a hybrid system with a unidirectional mmWave link using software burst ACK and well optimized packet and burst sizes can achieve competitive performance.

This paper presents a new type of wireless networking applications in data centers using steered-beam mmWave links. By taking advantage of clean LOS channels on top of server racks, robust wireless packet-switching network can be built. The transmission latency can be reduced by flexibly bridging adjacent rows of racks wirelessly without using long cables and multiple switches. Eliminating cables and switches also reduces equipment costs as well as server installation and reconfiguration costs. Security can be physically enhanced with controlled directivity and negligible wall penetration. The aggregate data transmission BW per given volume is expected to scale as the fourth power of carrier frequency. The paper also deals with the architecture of such network configurations and a preliminary demonstration system.

This paper describes a demonstration of multi-Gbps wireless systems using 60-GHz SiGe radio chipsets with noncoherent detection techniques. The modulation and demodulation are performed with BW-efficient frequency-shift keying schemes, such as MSK, by partitioning the function into the radio chip and baseband FPGA. With integrated mixer/sign modulator and appropriate precoder, an IQ-based radio interface is used in TX, while the noncoherent detections are achieved by using FM discriminator and CDR (Clock Data Recovery) in RX. As a result, multi-Gbps data rate systems can be built without using high sampling rate ADC/DAC, and the efficiency better than ASK or FSK modulation schemes is achieved.

We report a proof-of-concept demonstration of 2- Gbps uncompressed HDTV transmission using a 60-GHz SiGe radio chipset. We took a single-carrier approach with a usual DQPSK modulation scheme, assuming an LOS environment, and implemented the system with FPGAs. At the same time, in order to take care of more frequent sync/burst errors in high-data-rate single-carrier approaches, we equipped the baseband with effi- cient random/packet error recovery and symbol-timing recovery with an effective interpolation method. As a result, a clear and crisp image was obtained in the end-to-end transmission. I. INTRODUCTION

We present a multiple-packet recovery technique using a set of partially-overlapped nonbinary block codes. By reusing otherwise wasted information in the neighboring frames, it can recover a larger number of lost packets than the number of redundant packets in a frame, without increasing the asymptotic complexity of its decoding algorithm. A sample encoder/decoder pair is implemented as a soft IP to demonstrate a practical use of the proposed technique