Graduate Seminar: Saif Islam, Oct. 28, 2026

Summary

Professor Saif Islam presented work on making nanophotonic imaging and spectroscopy systems smaller, faster, and more sensitive by combining engineered semiconductor sensors with AI and machine-learning analysis. The goal is a low-cost system that can measure both the color (wavelength) and lifetime of emitted light on a CMOS chip.

Why measure color and lifetime?

• When illuminated, molecules can emit fluorescent light. The emitted wavelength helps identify a molecule, but different molecules—including healthy and cancer-related molecules—may emit similar colors.
• Measuring fluorescence lifetime—the delay between excitation and emission—provides a second identifying feature. Combining color and lifetime can distinguish molecules more reliably.
• The approach could support:
  • Cancer imaging: Identify tumor boundaries more precisely so a surgeon or robotic system can remove diseased tissue while sparing healthy tissue.
  • Agriculture: Monitor nutrient movement and uptake in plants in real time, potentially helping researchers adjust growing conditions to produce crops with desired mineral content.
  • Other molecular sensing and biomedical diagnostics.

The system should work with very little light. Intense illumination can damage or bleach tissue, while low-light measurements require high sensitivity and low noise. Real-time operation is also important for applications such as tracking molecules or guiding surgical tools.

The silicon detector challenge

Silicon is inexpensive and compatible with CMOS manufacturing, but it presents a trade-off:

• Thin silicon detectors can respond quickly because charge carriers travel short distances, but they absorb relatively little light—particularly at longer wavelengths.
• Thicker detectors absorb more light but collect carriers more slowly, limiting their speed.
• Other semiconductor materials can absorb light more effectively or respond faster, but can be more expensive and harder to integrate with silicon electronics.

Nanophotonic structures improve silicon sensing

Islam’s group uses carefully designed surface patterns, with dimensions related to the wavelength of light, to redirect incoming photons sideways through a thin silicon layer.

The basic strategy is to:

1. Pattern the silicon surface with wavelength-scale diffraction features.
2. Choose the pattern’s dimensions and spacing to couple incoming light into lateral optical modes.
3. Keep light traveling within the silicon for longer, increasing the chance that it will be absorbed while keeping the silicon thin enough for rapid carrier collection.
4. Passivate the patterned surface to reduce defects and noise caused by fabrication.

The speaker described this as making photons travel sideways through the material rather than simply passing through it. He cited surface texturing and nanophotonic methods as ways to increase absorption substantially. In the group’s reported results, patterned silicon detectors achieved absorption of about 40,000 per centimeter, compared with roughly 500 per centimeter for unstructured silicon in the cited example. A one-micron-thick device reportedly reached about 85% efficiency in its best result, and the group demonstrated suitability for 50-gigabit-per-second transmission.

AI-assisted wavelength sensing

Rather than using a bulky grating-based spectrometer to separate every wavelength, the group creates an array of detector pixels with deliberately different optical responses.

The approach is to:

1. Fabricate pixels with different surface-pattern dimensions or periodicities, so each responds differently to incoming wavelengths.
2. Calibrate the array using known light and record each pixel’s response.
3. Train an algorithm to map those response patterns to wavelengths or spectra.
4. Use the trained model to estimate the spectrum of unknown light from the array’s combined measurements.

Because the detectors are fast, the system can also measure fluorescence lifetime, enabling wavelength and lifetime sensing in the same compact platform. The speaker said that more pixels generally improve spectral resolution: arrays of 8, 16, and 32 sensors were compared, with 32 approaching the performance of a commercial spectrometer in the demonstrated results. However, high-speed packaging and interconnect requirements make larger arrays more difficult. The group considered 16 sensors adequate for its target applications.

Integration, applications, and limitations

• The long-term aim is a CMOS-integrated device containing the sensors and supporting electronics, such as amplifiers, transimpedance amplifiers, converters, and readout circuitry. A sufficiently small system could fit on the tip of an endoscope.
• Some designs included avalanche-photodiode-like internal gain, with reported gain from about 200 to 4,000. The specific application setup discussed in the Q&A used PIN photodiodes with transimpedance amplifiers, not the avalanche devices.
• Fabrication remains challenging. Surface patterning can create defects and leakage, so effective passivation is essential. University labs can make a wider variety of structures than commercial foundries, whose processes offer fewer patterning options. The group was testing foundry-manufactured devices; an initial attempt had failed, and a second batch was pending at the time of the talk.
• In response to questions, Islam said the sensors showed little angular dependence up to roughly 35–36 degrees. Beyond about 40 degrees, surface scattering became more significant. The proof-of-concept surgical experiments were conducted with other operating-room lights turned off to reduce background light.
• The speaker also emphasized that semiconductor manufacturing and workforce training are important broader opportunities, given the growth of semiconductor facilities and the range of materials used in the field.

Speakers and sources featured

Speakers

• Professor Saif Islam — seminar speaker, University of California, Davis.
• Dr. Kim — seminar introducer and moderator; first name is unclear in the subtitles.
• Unnamed audience members — asked questions during the discussion.

Researchers and research examples cited by the speaker

• Eli Yablonovitch and Sajeev John — associated in the talk with the photonic-crystal concept and light-trapping limits.
• Shanhui Fan — cited for nanophotonic methods that can enhance light absorption.
• Omid Momeni — cited for a high-speed silicon optical receiver demonstration.
• Laura Marks — described as a medical-school collaborator leading a center referred to in the subtitles as NCIBT.
• A University of Cambridge nanowire-spectrometer researcher — the name in the subtitles is unclear.
• Unnamed University of Wisconsin and Italian research groups — cited for alternative compact spectral-sensing and combined wavelength/lifetime demonstrations.