Paper

{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.12934v1</id>\n    <title>Photonic Exponential Approximation via Cascaded TFLN Microring Resonators toward Softmax</title>\n    <updated>2026-03-13T12:29:03Z</updated>\n    <link href='https://arxiv.org/abs/2603.12934v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.12934v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>The rapid growth of large-scale AI models has intensified energy consumption and data-movement challenges in modern datacenters.\n  Photonic accelerators offer a promising path by executing the linear matrix multiplications of transformer inference at high throughput\n  and low energy. However, the softmax attention layer -- which requires element-wise exponentiation followed by normalization -- still\n  relies on electronic post-processing, creating an electro-optic conversion bottleneck that negates much of the potential photonic\n  advantage.\n  We present a cascaded micro-ring resonator (MRR) architecture that synthesizes the per-channel exponential function required by softmax,\n  e^{x_n - max(x)}, over a finite interval with tunable worst-case relative error. A control signal detunes each ring via an\n  electro-optic mechanism; a weak probe at fixed frequency experiences Lorentzian transmission, and cascading N identical stages yields a\n  multiplicative transfer function whose logarithm is approximately linear.\n  We derive mapping rules, depth-scaling estimates, and a minimax fitting formulation, and validate the framework with three-dimensional\n  FDTD simulations of X-cut thin-film lithium niobate (TFLN) add-drop micro-ring resonators. Direct multi-ring FDTD validation extends to\n  a five-ring cascade and confirms agreement with theory primarily over the upper operating range; deeper cascades and higher quality\n  factors are assessed analytically. The cascade implements the per-channel exponential block -- the key missing nonlinearity for photonic\n  softmax; completing a full softmax additionally requires summation and normalization, which we discuss but do not implement here.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='physics.optics'/>\n    <published>2026-03-13T12:29:03Z</published>\n    <arxiv:comment>33 pages, 14 figures, includes supplementary material</arxiv:comment>\n    <arxiv:primary_category term='physics.optics'/>\n    <author>\n      <name>Hyoseok Park</name>\n      <arxiv:affiliation>Department of Physics, Chungnam National University, Daejeon, Republic of Korea</arxiv:affiliation>\n    </author>\n    <author>\n      <name>Yeonsang Park</name>\n      <arxiv:affiliation>Department of Physics, Chungnam National University, Daejeon, Republic of Korea</arxiv:affiliation>\n    </author>\n  </entry>"
}