Artificial intelligence is facilitating human life in many aspects. Previous artificial intelligence has been mainly focused on computer algorithms (e.g. deep-learning and extremelearning) and integrated circuits. Recently, all-optical diffractive deep neural networks (D 2 NN) were realized by using passive structures, which can perform complicated functions designed by computer-based neural networks at the light speed. However, once a passive D 2 NN architecture is fabricated, its function will be fixed. Here, we propose a programmable artificial intelligence machine (PAIM) that can execute various intellectual tasks by realizing hierarchical connections of brain neurons via a multi-layer digital-coding metasurface array. Integrated with two amplifier chips in each meta-atom, its transmission coefficient covers a dynamic range of 35 dB (from -40 dB to -5 dB), which is the basis to construct the reprogrammable physical layers of D 2 NN, in which the digital meta-atoms make the artificial neurons alive. We experimentally show that PAIM can handle various deep-learning tasks for wave sensing, including image classifications, mobile communication coder-decoder, and real-time multi-beam focusing. In particular, we propose a reinforcement learning algorithm for on-site learning and discrete optimization algorithm for digital coding, making PAIM have autonomous intelligence ability and perform self-learning tasks without the support of extra computer.

For the intelligence of metamaterials, the -sensing mechanism and programmable reaction units are two important components for self-recognition and -determination. However, their realization still face great challenges. Here, we propose a smart sensing metasurface to achieve self-defined functions in the framework of digital coding metamaterials. A sensing unit that can simultaneously process the sensing channel and realize phase-programmable capability is designed by integrating radio frequency (RF) power detector and PIN diodes. Four sensing units distributed on the metasurface aperture can detect the microwave incidences in the x- and y-polarizations, while the other elements can modulate the reflected phase patterns under the control of a field programmable gate array (FPGA). To validate the performance, three schemes containing six coding patterns are presented and simulated, after which two of them are measured, showing good agreements with designs. We envision that this work may motivate studies on smart metamaterials with high-level recognition and manipulation.

electromagnetic (EM) waves. Numerous interesting physical functions are realized by metamaterials [1][2][3][4][5] such as negative refraction, [1] invisibility cloaks, [2,3] and superlens. [4] Metasurfaces are planer versions of bulk metamaterials to control the EM waves on their phases, [6] amplitudes, [7] and polarizations. [8] Evolved from traditional metamaterials described by effective medium parameters, digital coding metamaterials have been proposed [9] to design metasurfaces from digital perspective and realize programmable functionalities using a single metasurface. In the digital coding metasurfaces, two distinct EM responses can be encoded as digits "1" and "0," [10,11] Then the related metasurface design is transformed into digital process. As a result, some groundbreaking researches like convolution operations [12] and information entropy [13] for metasurfaces have been presented, suggesting potential relations between the digital world and the physical world. Within this perspective, we can not only encode the phase responses, but also other EM characteristics like polarizations and orbital angular momentum (OAM) modes. [14,15] Consequently, the digitally programmable methods bring plentiful functions, such as space-time coding, [16,17] and new-architecture communication system. [18][19][20] As an important concept in microwave and optical frequencies, nonreciprocity has raised in myriad physical branchesthermodynamics, mechanics, electromagnetism, optics, and so on. [21] The classic nonreciprocal devices like circulators are indispensable components in radar [22] and communication systems. [23] For traditional electromagnetics, nonreciprocal devices are predominantly based on magnetized materials, [24,25] like ferrites, usually composed of iron oxides and other elements (Al, Ni, Co). [26] However, these kinds of materials are usually large, bulky, uneconomical, and difficult to integrate into the metasurfaces. To overcome these drawbacks, magnetless nonreciprocial metasurfaces (MNMs) have been proposed. [27][28][29][30][31] By interconnecting transistor-like amplifier [27][28][29] and isolator [30,31] into metasurface structure, MNMs can easily achieve smaller size and better integrability. In addition, nonreciprocial metamaterials based on time modulation have also been presented, [32] but their control system complexity is greatly increased.

Controlling the polarization state of electromagnetic waves is an important topic in microwaves due to the enormous application potential in radar technology and mobile communications. Here, we propose a programmable metasurface based on single-pole double-throw switches to realize multifunctional polarization conversions. A structure of the double-sided metallic pattern is adopted in the metasurface, in which a novel double-pole double-throw hub is achieved to guide the energy direction. Such a mechanism successfully induces multiple transmission channels into the metasurface structure for functional design. By controlling the states of the switches with a field programmable gate array, the x- and y-polarizations of the incident waves can be efficiently modulated into linear co- and cross-polarizations of transmitted waves, suggesting a higher degree of freedom on wave manipulations. The proposed metasurface can be developed as a near-field information encoder to transmit binary coding sequence according to the energy distribution. Character transmissions are realized by programming binary ASCII codes on the transmitted fields. Nine supercells on the metasurface can encode 9-bit binary information in one frame of near-field imaging, which can be switched in real time with high speed. We envision that this work will develop digital coding applications to control the polarization information.

No abstract