Inverse design of tunable electromagnetically induced transparency terahertz metamaterial antibiotic sensor by improved transformer

We propose a graphene-based tunable EIT-like terahertz metamaterial antibiotic sensor, achieving dynamic tunability of the EIT-like sensor and improving the transformer network to enable rapid reverse design of the sensor. This EIT-like sensor is excited by two cloud-shaped graphene resonant rings, which generate an excitation–excitation coupling mode, with the center frequency of its transparent window located at 1 THz of the chlortetracycline fingerprint spectrum, capable of achieving specific sensing detection of chlortetracycline, with a sensing sensitivity of 220 GHz/RIU, a Q value of 13.6, and an figure of merit of 9.84. The sensor can adjust the EIT-like window by changing the Fermi level of graphene by adjusting the applied voltage. Meanwhile, an improved transformer network architecture is proposed. According to the characteristics of EIT spectral data, random position encoding is used in the embedding module, the normalization layer is moved forward in the encoding module, and the decoding module is optimized to a fully connected layer. Eventually, the mean squared error of the validation set is 0.019, providing a solution for the on-demand design of metamaterial devices.