Taylor658/SiN-photonic-waveguide-loss-efficiency

💎 SiN Photonic Waveguide Loss & Efficiency Dataset

🔬 90,000 synthetic rows of silicon nitride (Si₃N₄) waveguide parameters linking geometry, fabrication, and operating conditions to loss and efficiency metrics, for regression modeling, simulation, and fine-tuning.

⚠️ Disclaimer: All rows are synthetically generated. Parameter ranges are informed by published SiN platform values, but no row is a foundry measurement. The data_source column is a schema field; every row in this release is labeled Synthetic.

---

📋 At a Glance

🔢 Rows	90,000
📊 Columns	25
🧪 Material	Silicon Nitride (Si₃N₄)
🎯 Primary Targets	propagation_loss, insertion_loss
🧬 Generation	Synthetic, parameterized over geometry, process, and operating ranges
🌍 Language	English
📄 License	MIT
🏷️ Version	1.1.1

---

✨ Key Highlights

• 💎 Material Focus: Silicon Nitride (Si₃N₄) on SiO₂ cladding
• 📊 25 Structured Columns: waveguide geometry, fabrication method, operating conditions, and synthetic performance metrics
• 📦 90,000 Rows: sufficient for training and validation splits
• 🎯 Use Cases:
  • 📏 Straight and bent waveguide loss prediction
  • 🏭 Process control and optimization studies
  • 📐 Photonic design parameter sensitivity review
  • 🔁 Synthetic data augmentation for smaller measured datasets

---

🏗️ Dataset Structure

Each row corresponds to a single waveguide configuration or measurement instance.

📐 Waveguide Geometry

Column	Unit	Description
waveguide_width	µm	Core width
waveguide_height	nm	Core thickness
bend_radius	µm	Bend radius for curved sections
device_length	mm	Total device length

🏭 Material & Fabrication

Column	Unit	Description
cladding_material	categorical	Cladding material (e.g., SiO₂)
cladding_thickness	µm	Cladding layer thickness
deposition_method	categorical	Core deposition process (e.g., LPCVD, PECVD)
etch_method	categorical	Etch process (e.g., RIE)
sidewall_roughness	nm	RMS sidewall roughness
annealing_params	string	Anneal temperature and duration (e.g., 900C_3hr)

⚙️ Operational Parameters

Column	Unit	Description
wavelength	nm	Operating wavelength
polarization	categorical	TE or TM
input_power	dBm	Launched optical power
temperature	°C	Device temperature

📈 Performance Metrics

Column	Unit	Description
insertion_loss	dB	Total device insertion loss
propagation_loss	dB/cm	Loss per unit length
coupling_efficiency_input	%	Input coupling efficiency
coupling_efficiency_output	%	Output coupling efficiency
scattering_loss	dB/cm	Scattering contribution to loss (independently sampled; not derived from sidewall_roughness)
effective_index	dimensionless	Modal effective index
mode_confinement_factor	0 to 1	Fraction of modal power in the core

🏷️ Metadata

Column	Unit	Description
batch_id	string	Fabrication batch or wafer identifier
data_source	categorical	Schema field for provenance; all rows in this release are Synthetic
measurement_method	categorical	Simulated measurement technique (e.g., cut-back, ring_resonance)
uncertainty	dB or %	Nominal uncertainty on the primary metric

---

🔍 Example Row

waveguide_width                = 1.212
waveguide_height               = 400.00
cladding_material              = SiO2
cladding_thickness             = 2.50
deposition_method              = LPCVD
etch_method                    = RIE
sidewall_roughness             = 2.05
annealing_params               = 900C_3hr
wavelength                     = 1452.23
polarization                   = TE
input_power                    = 0.00
temperature                    = 25.00
bend_radius                    = 50.00
device_length                  = 10.00
insertion_loss                 = 3.50
propagation_loss               = 0.300
coupling_efficiency_input      = 72.00
coupling_efficiency_output     = 68.00
scattering_loss                = 0.15
effective_index                = 1.800
mode_confinement_factor        = 0.80
batch_id                       = BATCH_12
data_source                    = Synthetic
measurement_method             = cut-back
uncertainty                    = 0.05

---

💡 Usage

🚀 Load the Dataset

from datasets import load_dataset

dataset = load_dataset("Taylor658/SiN-photonic-waveguide-loss-efficiency", split="train")
print(dataset[0])

🐼 Explore with pandas

import pandas as pd

df = dataset.to_pandas()
print(df.describe())

🎯 Regression Example

Treat geometry, fabrication, and operating columns as features and propagation_loss or insertion_loss as the target:

features = df[[
    "waveguide_width", "waveguide_height", "sidewall_roughness",
    "wavelength", "polarization", "temperature"
]]
target = df["propagation_loss"]

# Encode categorical columns, then fit a regression model (scikit-learn, XGBoost, etc.)

🔗 Companion Resources

• 🛠️ Photonic-Waveguide: agentic SiN waveguide design workflow (modesolverpy, SAX+JAX, gdsfactory via FastMCP)
• 📊 photonic-integrated-circuit-yield: companion yield dataset
• 🤖 Photonics_Distill_Llama_70B: photonics reasoning model

---

⚠️ Limitations

🧬 Synthetic Content

• Parameter ranges are informed by published SiN platform values, but actual values vary by foundry, tool set, and process
• No row is a measured device

📏 Physics Simplifications

• Not all real-world correlations are captured; effects such as stress-induced birefringence, N-H absorption near 1520 nm, and wavelength-dependent index dispersion may not be modeled
• scattering_loss and sidewall_roughness are independently sampled; the physical dependence between them is not modeled

📐 Measurement Realism

• The uncertainty column is a nominal value and does not reproduce real measurement artifacts (facet variation, alignment drift, detector noise)
• measurement_method is a label; no measurement was performed

🚫 Not for Process Qualification

• Predictions trained on this data must be validated against measured devices before informing fabrication decisions

---

📄 License

Released under the MIT License. You may modify, distribute, and use it for commercial or non-commercial purposes with attribution.

---

🙌 Citation

@misc{sin_waveguide_loss_efficiency_2026,
  title  = {SiN Photonic Waveguide Loss \& Efficiency Dataset (Synthetic)},
  author = {Taylor, A.},
  year   = {2026},
  howpublished = {\url{https://huggingface.co/datasets/Taylor658/SiN-photonic-waveguide-loss-efficiency}}
}

---

🧑‍💻 Contributing

• 💬 Open a Discussion to propose additional process parameters, cladding materials, or loss mechanisms

---

👨‍🚀 Author: A Taylor · 🤗 Hugging Face: hf.co/Taylor658 · 🐙 GitHub: ATaylorAerospace