Spiking Photonic Neurons Based on Two-Section InP Quantum-Well Lasers Integrated on Silicon

Adonis Bogris; Benoit Charbonnier; Charis Mesaritakis; Menelaos Skontranis; Olivier Girard

REVIEW 3 major objections 5 minor 23 references

Spiking Photonic Neurons Based on Two-Section InP Quantum-Well Lasers Integrated on Silicon

T0 review · 3 major / 5 minor · reviewed 2026-07-30 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The same two-section InP laser on silicon can be switched by DC bias alone between resonate-and-fire and integrate-and-fire spiking, with design knobs that set how fast and how wide that integrate-and-fire window is.

desk verdict Solid first maps of dual bias-tunable spiking on III–V-on-silicon two-section lasers; Regime III is well evidenced, Regime I’s resonate-and-fire label is thinner. read the letter →

arxiv 2607.26950 v1 pith:ZI7GAVOS submitted 2026-07-29 physics.optics

Menelaos Skontranis , Benoit Charbonnier , Olivier Girard , Adonis Bogris , Charis Mesaritakis This is my paper · ORCID

classification physics.optics

keywords photonicneuronsspikinglaserstwo-sectionInPonsiliconintegrate-and-fireresonate-and-fireneuromorphicphotonicsQ-switching

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper shows that two-section InP quantum-well lasers built monolithically on silicon can act as versatile high-speed photonic neurons. By changing only the gain current and the reverse bias on the saturable absorber, one physical device produces two distinct spiking styles: a slow, nearly fixed-rate resonate-and-fire-like mode and a fast integrate-and-fire-like mode whose pulse rate rises continuously with drive, up to about 1.8 GHz. The fast mode also shows a clear electrical threshold, temporal summation of subthreshold pulses, and a refractory period. Mapping nine cavity designs further shows that longer cavities and moderate absorber fractions enlarge the fast-spiking bias window and raise the maximum pulse rate. The practical point is that CMOS-compatible laser neurons can sit on the same chip as low-loss silicon waveguides that would serve as synapses, without needing separate device types for different neural primitives.

What carries the argument

Two-section gain–saturable-absorber laser dynamics under independent electrical bias, partitioned into self-pulsation Regime I (linked by the authors to optothermal Canard spikes) and Regime III (linked to gain–SA Q-switching). The design parameters cavity length L and absorber-length ratio RABS set the size of the Regime III operating window and the highest achievable pulse rate.

What would settle it

Measure temperature, carrier, or small-signal dynamics on the same devices: if the slow ~16 MHz pulses lack a thermal time-scale signature, or if the fast pulses fail threshold/integration/refractory tests under optical rather than electrical drive, the claimed neural-regime mapping collapses.

Watch

Extended reading notes

Core claim

A single two-section InP quantum-well laser monolithically integrated on silicon realizes both resonate-and-fire-like and integrate-and-fire-like spiking regimes through DC bias alone. Regime I (low current, low absorber bias) yields broad pulses at a nearly constant ~16 MHz rate whose amplitude grows with current; Regime III (higher current and absorber bias) yields sub-nanosecond pulses whose rate rises continuously to ~1.8 GHz while amplitude saturates. Regime III further exhibits threshold, temporal integration, and refractory recovery under electrical perturbation. Cavity length and gain/absorber length ratio systematically expand or shrink the Regime III bias region and its maximum pul

Load-bearing premise

The paper treats the shape of pulse rate versus current, plus analogy to earlier literature, as enough to identify the slow regime with optothermal Canard dynamics and the fast regime with ordinary gain–absorber Q-switching, without separate thermal or carrier measurements on this silicon platform.

Editorial extensions

If this is right

One laser geometry can supply both resonate-and-fire and integrate-and-fire primitives simply by changing DC bias, reducing the need for heterogeneous neuron types on a neuromorphic PIC.
Choosing L around 800 µm and RABS near 5% maximizes the usable integrate-and-fire bias window and pulse rate (up to GHz-class) for time-coded photonic networks.
Monolithic III–V-on-silicon lasers can sit beside low-loss silicon waveguide synapses on the same chip, supporting denser spiking photonic circuits than hybrid assembly.
Spike duration and refractory time set by these lasers bound the temporal resolution and processing latency of any network that uses them as neurons.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Because both regimes live in one device, a network could reconfigure neuron type at run time by bias, enabling mixed integrate/resonate architectures without redesigning the chip.
The reported electrical refractory window (~0.55–0.8 ns) implies that all-optical synaptic summing on the same platform would need to respect sub-nanosecond recovery if GHz rates are to be used.
If silicon waveguide delay lines are later used as synapses, the slower Regime I (~16 MHz) may be easier to match with longer on-chip delays than the GHz Regime III.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The real addition here is experimental, not new laser physics: nine geometries of two-section InP QW lasers on silicon, regime maps vs I and VABS, and clear L/RABS trends for the fast pulsation window and max PRF. That platform-plus-design-rule package is useful for people building neuromorphic PICs who actually care about monolithic III–V on Si rather than hybrid III–V cavities.

What they do well is the Regime III story. PRF rising with current to ~1.8 GHz, amplitude saturation, and the four electrical-pulse checks (threshold, two-pulse summation, refractory recovery) are direct and match the integrate-and-fire claim. The systematic maps (RABS ≤5% preferred; longer L widens the window and raises PRF) are the part I would actually use. Circularity is low—these are measured traces and bias maps, not model-forced outputs. Citations to the classic Q-switching and laser-neuron literature are appropriate.

The soft spot is real but scoped. The dual-neuron versatility claim needs both regimes. Regime I is only free-running: ~16 MHz PRF almost flat in current, amplitude rising, plus Canard/optothermal analogy. No pulsed drive, resonance curve, or phase-preference test. Constant slow PRF is compatible with several mechanisms; calling it resonate-and-fire is a literature label, not a demonstrated I/O property on this chip. That weakens the “same device, both neuron types by DC bias alone” line without killing the paper. Minor gaps: no error bars/repeatability, closed data, low external efficiency from the Sagnac facets, and abstract language about scalable neuromorphic PICs that outruns single-laser evidence.

Who it’s for: hardware groups doing spiking lasers and Si photonic integration. Not for theorists hunting new dynamics. Math is not load-bearing; data and citation pattern look honest. I would send it to referees—experimental letter with a clear platform result and one claim that needs tightening. Engage if you care about III–V-on-Si neurons; skim the maps if you only need the design rules.

Referee Report

3 major / 5 minor

Summary. The manuscript experimentally characterizes spiking dynamics in two-section InP quantum-well lasers monolithically integrated on silicon. By tuning gain current and SA reverse bias, the authors map three self-pulsation regimes and identify Regime I (nearly constant ~16 MHz PRF, amplitude rising with current) with resonate-and-fire-like behavior and Regime III (PRF rising continuously to ~1.8 GHz, amplitude saturating) with integrate-and-fire-like behavior. For Regime III they further demonstrate electrical-pulse threshold, two-pulse temporal integration, and a refractory period. Systematic regime maps for nine geometries (L = 400–800 µm, RABS = 3–10%) show how cavity length and absorber ratio expand or shrink the Regime III bias window and maximum PRF. The work positions these CMOS-compatible lasers as versatile spiking neurons for neuromorphic PICs with silicon waveguide synapses.

Significance. If the dual-regime and design-parameter claims hold, the paper is a useful experimental step for neuromorphic silicon photonics: it shows that monolithically integrated III–V-on-Si two-section lasers can support GHz-class integrate-and-fire-like spiking and that L and RABS systematically control the usable bias space and PRF. The Regime III excitability suite (threshold, summation, refractory) and the nine-device regime maps are concrete, reusable data for device design. The platform argument—active laser neurons coexisting with low-loss Si interconnects—is timely. Credit is due for the breadth of the geometry survey and for grounding the IF claim in direct pulsed-stimulus tests rather than free-running traces alone. The main significance risk is over-identification of Regime I as a resonate-and-fire computational primitive without matching I/O tests; if that label is softened or supported, the remaining IF-plus-design contribution remains solid for a letter-length optics/photonics venue.

major comments (3)

[Abstract; Figs. 2a,b vs Fig. 3; discussion of Regimes I and III] Abstract, introduction, and conclusion present the same device as realizing both integrate-and-fire and resonate-and-fire neurons via DC bias alone. Regime III is supported by PRF-vs-current shape (Fig. 2c,d) and by direct electrical tests of threshold, temporal integration, and refractory recovery (Fig. 3). Regime I is supported only by free-running traces: nearly constant ~16 MHz PRF and amplitude rising with current (Fig. 2a,b), plus analogy to Canard/optothermal pulsations [10] and the PRF-shape taxonomy in [8,9]. Constant free-running PRF is compatible with several slow self-pulsation mechanisms and does not by itself establish resonator-neuron computational behavior (preferred-frequency response, phase-dependent firing, or pulsed-stimulus resonance). Either (i) add a minimal Regime I stimulus test (e.g., weak periodic or paired electrical perturbations showing frequency/phase prefe
[Paragraphs linking Figs. 2a–d to refs. [8–11]] Mechanism assignments for both regimes rest on literature analogy (Canard/optothermal for Regime I; gain–SA Q-switching for Regime III) without independent thermal, carrier, or small-signal measurements on this Si-integrated platform. That is acceptable as interpretation if labeled as such, but the text currently states the mechanisms as established (“linked to Canard spikes”, “attributed to gain–SA and coupled dynamics”). Please separate observation from interpretation: report the measured PRF/amplitude/excitability facts first, then discuss candidate mechanisms and what would distinguish them on this platform (e.g., thermal time-scale estimates, temperature dependence, or small-signal response).
[Fig. 2; Fig. 4 and associated text on L and RABS] Regime maps (Fig. 4) and PRF/PP trends (Fig. 2) are central to the design-parameter claims (L expands Regime III and raises max PRF; RABS ≲ 5% preferred; highest PRF at RABS = 5% not 3%). No repeatability, device-to-device spread, or uncertainty is reported for regime boundaries or for the “X” max-PRF points. For a systematic nine-geometry study, at least indicate whether maps are single-shot or repeated, and give uncertainty or range on the quoted maximum PRFs (including the 1.8 GHz figure). Without that, the quantitative design guidelines are harder to trust for follow-on PIC design.

minor comments (5)

[Fig. 3 caption vs body text] Fig. 3 caption states the suprathreshold pair separation in (c) as 0.4 ns, while the main text says 0.5 ns for the case that yields a single spike. Align caption and text.
[Fig. 1c and surrounding text] Fig. 1c grey/yellow/orange/red regime map is described in the caption but the criteria used to assign a bias point to Regime I vs II vs III (e.g., FWHM thresholds, presence of burst coexistence) are only qualitative in the text. A short explicit classification rule would help reproducibility.
[Initial characterization paragraph; Fig. 2b,d] External quantum efficiency and coupling losses (6–7 dB) are noted; on-chip or facet-corrected powers would make peak-power comparisons across geometries more meaningful if space allows.
[Throughout; References] Typographical/formatting: “VABS” / “ABS” / “RABS” notation is slightly inconsistent; “In this work we” opening is duplicated between abstract-like front matter and body; some references appear twice (numbered list plus expanded list).
[Device/PIC description] Sagnac loop reflectivities (10% and 40%) are given as design values; a brief note on whether they were verified would strengthen the cavity description.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: experimental bias maps and time traces stand alone; neuron-type labels are interpretive analogies, not inputs that force the measured PRFs.

full rationale

The paper’s load-bearing content is measured optical power, free-running pulse trains, PRF/peak-power vs bias, and electrical-pulse excitability tests on fabricated two-section InP-on-Si lasers, plus systematic maps over cavity length L and absorber ratio RABS. Regime boundaries and PRF values are read from oscilloscope data, not derived from a fitted dynamical model that is then re-presented as prediction. Literature citations ([8–11]) supply only interpretive names (resonate-and-fire / Canard; integrate-and-fire / Q-switching) for already-observed PRF-vs-current shapes; they do not algebraically produce the reported frequencies or the Fig. 3 threshold/integration/refractory traces. There is no self-citation chain among the present authors that underwrites uniqueness or forbids alternatives, no parameter fitted to one subset and “predicted” on a closely related subset, and no renaming of a known empirical law as a first-principles result. Any weakness in the Regime I resonate-and-fire identification is an evidence/correctness issue, not circularity. Score 0 is therefore appropriate.

Assumptions & free parameters 2 free parameters · 4 assumptions · 1 invented entities

Load-bearing content is experimental observation plus standard semiconductor-laser domain knowledge. No fitted global theory constants drive the central claim. Interpretive axioms import Canard/optothermal and Q-switching pictures from the literature to name the regimes; design parameters L and RABS are swept experimentally rather than fitted. No new physical entities are postulated.

free parameters (2)

Operating bias points (I, VABS) chosen near regime boundaries for excitability tests = I=71.24 mA, VABS=-2.2 V (example point)
I=71.24 mA and VABS=-2.2 V are hand-selected near the Regime III boundary so that 0.3 V vs 0.8 V pulses demonstrate threshold/integration; different nearby biases could change apparent thresholds.
Sagnac loop facet reflectivities = 10% and 40%
Fixed at 10% and 40% by design; they set threshold and external efficiency and therefore the absolute bias windows, but are not fitted to spiking data.

assumptions (4)

domain assumption Nearly current-independent PRF with rising amplitude indicates resonate-and-fire / Canard (optothermal) dynamics; strongly current-dependent PRF with saturating amplitude indicates integrate-and-fire / Q-switching gain–SA dynamics.
Used to label Regimes I and III via refs. [8–11] without independent thermal or carrier diagnostics on these devices.
domain assumption Standard two-section laser rate-equation picture: pulse formation from interplay of gain recovery and intensity-dependent SA bleaching.
Background for interpreting Regime III and the effect of longer SA (higher saturation energy) on bias windows; drawn from [6,11,12].
domain assumption Electrical perturbations on the absorber section are a valid proxy for testing neuronal threshold, integration, and refractoriness of the optical spike generator.
Underpins the Fig. 3 isomorphism claims; common in the laser-neuron literature [8] but is a modeling choice about what counts as a synaptic input.
domain assumption Temperature held at 25 °C and probe/fiber coupling are stable enough that observed regimes are intrinsic device dynamics.
Stated characterization conditions; unquantified drift would blur regime boundaries.

invented entities (1)

Regimes I / II / III classification
purpose: Organize observed self-pulsation behaviors by PRF, FWHM, and peak-power phenomenology on the bias map.
Observational labels local to this paper’s maps, not new physical objects; Regime II (burst) is introduced then de-emphasized.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Spiking Photonic Neurons Based on Two-Section InP Quantum-Well Lasers Integrated on Silicon." pith.science (2026). https://pith.science/paper/ZI7GAVOS

@misc{pith2026260726950,
  author       = {Pith},
  title        = {Pith review of: Spiking Photonic Neurons Based on Two-Section InP Quantum-Well Lasers Integrated on Silicon},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZI7GAVOS}},
  note         = {Machine review of arXiv:2607.26950}
}

read the original abstract

In this work we experimentally investigate the spiking dynamics of two-section InP quantum-well lasers monolithically integrated on silicon. By appropriately tuning the electrical bias conditions, we realize multiple neuronal-like operating regimes, such as integrate-and-fire and resonate-and-fire, highlighting the device's versatility as a high-speed photonic neuron. A systematic investigation of laser design parameters, including cavity length and gain/saturable absorber ratio, elucidates their impact on spiking-related properties (such as pulse repetition frequency) and traces the operational parameter space that unlocks stable spiking. Finally, these findings pave the way toward scalable neuromorphic photonic integrated circuits, where low-loss silicon synapses coexist with versatile laser neurons.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references

[10]

Tierno, N

A. Tierno, N. Radwell, and T. Ackemann, Phys. Rev. A 84, 043828 (2011)

2011
[1]

Mitchell Waldrop, Nature 530, 144 (2016)

2016
[2]

Marković, A

D. Marković, A. Mizrahi, D. Querlioz, et al, Nat Rev Phys 2, 499 (2020)

2020
[3]

B. J. Shastri, A. N. Tait, T. Ferreira de Lima, W. H. P. Pernice, H. Bhaskaran, C. D. Wright, and P. R. Prucnal, Nat. Photonics 15, 102 (2021)

2021
[4]

Shekhar, W

S. Shekhar, W. Bogaerts, L. Chrostowski, J. E. Bowers, M. Hochberg, R. Soref, and B. J. Shastri, Nat Commun 15, 751 (2024)

2024
[5]

Farmakidis, B

N. Farmakidis, B. Dong, and H. Bhaskaran, Nat Rev Electr Eng 1, 358 (2024)

2024
[6]

G. J. Spühler, R. Paschotta, R. Fluck, B. Braun, M. Moser, G. Zhang, E. Gini, and U. Keller, J. Opt. Soc. Am. B 16, 376 (1999)

1999
[7]

Uenohara, R

H. Uenohara, R. Takahashi, Y. Kawamura, and H. Iwamura, IEEE J. Quantum Electron. 32, 873 (1996)

1996

Show all 23 references

[8]

P. R. Prucnal, B. J. Shastri, T. Ferreira De Lima, M. A. Nahmias, and A. N. Tait, Adv. Opt. Photon. 8, 228 (2016)

2016
[9]

E. M. Izhikevich, Computational Neuroscience (MIT press, 2007)

2007
[12]

V. Z. Tronciu, M. Yamada, T. Ohno, S. Ito, T. Kawakami, and M. Taneya, IEEE J. Quantum Electron. 39, 1509 (2003). References Fig. 4 Map regimes for RABS=3% and L=400, 600 and 800μm (a-c), for RABS=5% and L=400, 600 and 800μm (d-f) and for RABS=8(10)% and L=400 and 800 (600)μm ...

2003
[13]

Waldrop, M. M. (2016). More than Moore. Nature, 530(7589), 144-148

2016
[14]

Marković, D., Mizrahi, A., Querlioz, D., & Grollier, J. (2020). Physics for neuromorphic computing. Nature Reviews Physics, 2(9), 499-510

2020
[15]

J., Tait, A

Shastri, B. J., Tait, A. N., Ferreira de Lima, T., Pernice, W. H., Bhaskaran, H., Wright, C. D., & Prucnal, P. R. (2021). Photonics for artificial intelligence and neuromorphic computing. Nature Photonics, 15(2), 102-114

2021
[16]

E., Hochberg, M., Soref, R., & Shastri, B

Shekhar, S., Bogaerts, W., Chrostowski, L., Bowers, J. E., Hochberg, M., Soref, R., & Shastri, B. J. (2024). Roadmapping the next generation of silicon photonics. Nature Communications, 15(1),

2024
[17]

Farmakidis, N., Dong, B., & Bhaskaran, H. (2024). Integrated photonic neuromorphic computing: opportunities and challenges. Nature Reviews Electrical Engineering, 1(6), 358-373

2024
[18]

J., Paschotta, R., Fluck, R., Braun, B., Moser, M., Zhang, G.,

Spühler, G. J., Paschotta, R., Fluck, R., Braun, B., Moser, M., Zhang, G., ... & Keller, U. (1999). Experimentally confirmed design guidelines for passively Q-switched microchip lasers using semiconductor saturable absorbers. Journal of the Optical Society of America B, 16(3), 376-388

1999
[19]

Uenohara, H., Takahashi, R., Kawamura, Y., & Iwamura, H. (2002). Static and dynamic response of multiple-quantum-well voltage-controlled bistable laser diodes. IEEE journal of quantum electronics, 32(5), 873-883

2002
[20]

R., Shastri, B

Prucnal, P. R., Shastri, B. J., Ferreira de Lima, T., Nahmias, M. A., & Tait, A. N. (2016). Recent progress in semiconductor excitable lasers for photonic spike processing. Advances in Optics and Photonics, 8(2), 228-299

2016
[21]

Izhikevich, E. M. (2007). Dynamical systems in neuroscience. MIT press

2007
[22]

Tierno, A., Radwell, N., & Ackemann, T. (2011). Low-frequency self-pulsing in single-section quantum-dot laser diodes and its relation to optothermal pulsations. Physical Review A—Atomic, Molecular, and Optical Physics, 84(4), 043828

2011
[23]

Rachinskii, A

D. Rachinskii, A. Vladimirov, U. Bandelow, B. Hüttl, and R. Kaiser, J. Opt. Soc. Am. B 23, 663 (2006)

2006
[24]

Z., Yamada, M., Ohno, T., Ito, S., Kawakami, T., & Taneya, M

Tronciu, V. Z., Yamada, M., Ohno, T., Ito, S., Kawakami, T., & Taneya, M. (2003). Self-pulsation in an InGaN laser-theory and experiment. IEEE journal of quantum electronics, 39(12), 1509-1514

2003

Pith tools

Reviewed July 30, 2026 · model on record in the stance chip above.

[1] [10]

Tierno, N

A. Tierno, N. Radwell, and T. Ackemann, Phys. Rev. A 84, 043828 (2011)

2011

[2] [1]

Mitchell Waldrop, Nature 530, 144 (2016)

2016

[3] [2]

Marković, A

D. Marković, A. Mizrahi, D. Querlioz, et al, Nat Rev Phys 2, 499 (2020)

2020

[4] [3]

B. J. Shastri, A. N. Tait, T. Ferreira de Lima, W. H. P. Pernice, H. Bhaskaran, C. D. Wright, and P. R. Prucnal, Nat. Photonics 15, 102 (2021)

2021

[5] [4]

Shekhar, W

S. Shekhar, W. Bogaerts, L. Chrostowski, J. E. Bowers, M. Hochberg, R. Soref, and B. J. Shastri, Nat Commun 15, 751 (2024)

2024

[6] [5]

Farmakidis, B

N. Farmakidis, B. Dong, and H. Bhaskaran, Nat Rev Electr Eng 1, 358 (2024)

2024

[7] [6]

G. J. Spühler, R. Paschotta, R. Fluck, B. Braun, M. Moser, G. Zhang, E. Gini, and U. Keller, J. Opt. Soc. Am. B 16, 376 (1999)

1999

[8] [7]

Uenohara, R

H. Uenohara, R. Takahashi, Y. Kawamura, and H. Iwamura, IEEE J. Quantum Electron. 32, 873 (1996)

1996