Ultrafast neural sampling with spiking nanolasers
Ivan K. Boikov, Alfredo de Rossi, Mihai A. Petrovici · 2025
Read Ultrafast neural sampling with spiking nanolasers by Ivan K. Boikov, Alfredo de Rossi, Mihai A. Petrovici (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-14446.txt.
Introduction
Ivan Boikov, Alfredo de Rossi, and Mihai Petrovici show in theory in 2025 how networks of tiny spiking lasers could draw samples for probability based reasoning at very high speed. Optical computing draws interest for its speed, low power, and low delay against electronic chips. Small crystal lasers that fire brief light pulses, like nerve spikes, were lately shown to exist.
The team derives rules that translate a classic sampling network, the Boltzmann machine, into a network of these photonic neurons. In tests on generative tasks, the laser network sampled from learned distributions as designed. Each laser spike takes about a billionth of a second, so sampling could run far faster than on digital hardware. The authors also sketch what a real chip would need. Fast sampling serves speech, vision, and risk models.
Catalog record
| CREATOR | Ivan K. Boikov, Alfredo de Rossi, Mihai A. Petrovici |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 16 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-14446
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Ultrafast neural sampling with spiking nanolasers Ivan K. Boikov, Alfredo de Rossi, Mihai A. Petrovici Abstract Abstract Owing to their significant advantages in terms of bandwidth, power efficiency and especially speed, optical neuromorphic systems have arisen as interesting alternatives to conventional semiconductor devices. Recently, photonic crystal nanolasers with excitable behaviour were first demonstrated. Depending on the pumping strength, they emit short optical pulses – spikes – at various intervals on a nanosecond timescale. In this theoretical work, we show how networks of such photonic spiking neurons can be used for Bayesian inference through sampling from learned probability distributions. We provide a detailed derivation of translation rules from conventional sampling networks such as Boltzmann machines to photonic spiking networks and demonstrate their functionality across a range of generative tasks. Finally, we provide estimates of processing speed and power consumption, for which we expect improvements of several orders of magnitude over current state-of-the-art neuromorphic systems. Introduction In the overwhelming majority of artificial neural networks (ANNs) in use today, neuronal outputs are determined by smooth and continuous activation functions, and their values are updated synchronously for multiple neurons within a network. In contrast, biological neurons, especially in the mammalian cortex, communicate asynchronously with short, stereotypical pulses called “spikes” . While the non-differentiability of such spike-based codes indeed makes them more difficult to train using error backpropagation and therefore less attractive for conventional deep learning (however, see [ goltz2021fast , wunderlich2021event , stanojevic2022exact ] ), they possess several properties that are extremely relevant in physical neuronal networks, biological and artificial alike. First, they can be much more efficient than time-continuous or rate codes: as information becomes implicitly encoded in the timing of the spikes, rather than the value of neuronal outputs, a spike code can save both energy and bandwidth. Thus, spiking neurons have become the de-facto standard model for neuromorphic systems, both purely digital [ merolla2014 , furber2014spinnaker , davies2018loihi ] and mixed-signal [ pfeil2013six , moradi2017scalable , neckar2018braindrop , billaudelle2020versatile ] . Second, due to the all-or-nothing nature of spikes, the transmission of information is significantly less prone to disruption by noise. This makes spiking neural networks particularly attractive for mixed-signal devices, whose analog components (neurons and synapses) invariably introduce spatio-temporal noise into the network dynamics. By associating neuronal spikes with binary states, one can link spiking neural networks to Boltzmann machines (BMs) [ buesing2011neural , petrovici2013stochastic , petrovici2016stochastic ] . This machine learning model forms the basis for powerful variants such as deep belief networks [ hinton2006fast ] and autoencoders [ hinton2006reducing ] , which have been used for applications ranging from acoustic modeling [ mohamed2011acoustic ] to medical diagnostics [ abdel2016breast ] and quantum tomography [ carleo2017solving ] . By moving to the spiking domain, one can harness the speed and efficiency of neuromorphic substrates for such applications. Indeed, first demonstrations of such spike-based sampling (SBS) networks on highly accelerated neuromorphic hardware have already shown promising results [ kungl2019accelerated , czischek2022spiking , klassert2022variational ] . However, despite the impressive results of electric and electronic networks, electrical interconnects between computing elements limit bandwidth, latency and energy efficiency [ ho2001 ] . In contrast, photonic transmission of information offers significant advantages in terms of loss and speed. Not only is the issue of Joule heating alleviated, but the improved fan-in/-out by wavelength multiplexing also allows large-bandwidth light-speed transmission of signals. A photonic equivalent of a neuron is a laser with an excitable response [ wunsche2001 , romeira2013 , prucnal2016 ] ; interaction between such neurons with highly accelerated intrinsic dynamics is being developed [ robertson2019 ] . Among a variety of possible implementations, a semiconductor laser with a saturable absorber (SA) is of particular interest here [ barbay2011 ] because of the strong analogy with the leaky integrate-and-fire (LIF) neuron model [ nahmias2013 ] . Semiconductor lasers can be miniaturized and integrated in a photonic circuit and need tiny amounts of power (about 100 µW) for operation [ crosnier2017 ] . These are essential requirements for advancing the neuromorphic state of the art. Very recently, this technology was shown to produce an excitable response by introducing a section behaving as an SA [ delmulle2023 ] , making it promising for the implementation of efficient integrated photonic spiking neurons (PSNs). In this work, we demonstrate the capability of such PSN networks to implement BMs through SBS. Results Preliminaries Whenever a neuron fires a spike, it enters a refractory period of some duration \tau during which it cannot spike again. Thus, a spike can be interpreted as representing the onset of an “active” state z_{k}(t)=1\Leftrightarrow\text{neuron has fired in }(t-\tau;t]\;, (1) with z_{k}(t)=0 otherwise. In BMs, the probability of neuron ensemble being in a particular state \boldsymbol{z} follows the Boltzmann distribution p(\boldsymbol{z})\propto\exp\left[-E(\boldsymbol{z})/k_{\text{B}}T\right]\;, where the energy E is defined as E=-\sum_{k<j}W_{kj}z_{k}z_{j}-\sum_{k}b_{k}z_{k}\;, with neuronal biases b_{k} and synaptic weights W_{kj}=W_{jk} , k_{\text{B}} is the Boltzmann constant, and T is the ensemble (Boltzmann) temperature. Here, we disregard the units, assume k_{\text{B}}T=1 and omit this term for brevity. In SBS, a spiking neural network approximates p(\boldsymbol{z}) through a time-continuous analogon of Gibbs sampling [ buesing2011neural ] . A particularly interesting variant builds on LIF sampling (LIFS) neurons [ petrovici2013stochastic , petrovici2016stochastic ] , as they represent a de-facto standard model across the vast majority of neuromorphic platforms. In LIFS, the required stochasticity is generated by adding (or exploiting pre-existing) noise on the neuronal membranes, which consequently follow the Ornstein-Uhlenbeck process \tau_{\text{m}}\dot{u}_{k}=(b-u)+\sum_{j}\sum_{t^{\text{spike}}_{j}}W_{kj}\kappa(t-t^{\text{spike}}_{j})+\sigma_{W}\mathinner{\mathrm{d}}\mathchoice{\mskip-3.0mu}{\mskip-3.0mu}{}{}{}W (2) with an autocorrelation determined by the neuronal membrane time constant \tau_{\text{m}} ( \mathinner{\mathrm{d}}\mathchoice{\mskip-3.0mu}{\mskip-3.0mu}{}{}{}W represents a Wiener process scaled by a noise amplitude \sigma_{W} ). The interaction between neurons is mediated by an additive postsynaptic potential (PSP) kernel \kappa that is triggered by an incoming spike at time t^{\text{spike}} . Spiking nanolasers The PSNs discussed in this work are semiconductor lasers composed of two sections: a gain section and an SA. The two sections form a single resonator and are therefore spanned by a single laser mode ( Fig. 1 a). The gain section is pumped to reach local electron population inversion and stimulated emission of photons. The SA is not pumped, such that here absorption always dominates, yet saturating as the flux of absorbed photons increases. As the pump rate increases beyond a certain threshold, the absorption is overtaken by the stimulated emission; akin to negative differential resistance in electronic oscillators, this introduces a positive feedback which initiates the emission of a pulse. In turn, this depletes the population of excited electron-hole pairs, hence the gain, and the laser is shut off. Therefore, the emission of a new pulse immediately following a previous one is strongly suppressed. After some time, the pumping re-establishes the initial gain, and the nanolaser can spike again. This process ( Fig. 1 b) bears a strong resemblance to the generation of action potentials and the subsequent refractoriness found in biological neurons, as described by the Hodgkin-Huxley model [ hodgkin1952quantitative ] . Figure 1: Spiking nanolasers. (a) Schematic of a PSN coupled to a waveguide (gray). Here, the laser is a photonic crystal; its modes are standing-wave and optical spikes are coupled out to the waveguide in both directions, E is an electric field, \lambda is the wavelength in vacuum, and n is