Projects

EECE 530 Electro-Optics · Binghamton University · MATLAB

Design and analysis of the key components in a fiber-optic link: launching light into multimode fiber, handling random polarization, modulating a binary signal, and recovering it at a photodiode. Four linked tasks that build into one end-to-end system.

Task 1 — Fiber coupler

Single-lens system launching a laser into 50 µm multimode fiber (n₁ = 1.5, n₂ = 1.49, lens F = 3.5 mm, D = 2 mm).

Efficiency is limited because the lens focuses rays across a wider angular spread than the fiber's acceptance cone will admit.

Task 2 — Revised three-lens coupler

Redesigned with a three-lens arrangement that shrinks the beam so every ray arrives within θ_max, recovering the lost power.

Task 3 — Random polarization through a polarizer

Malus's law applied to a randomly polarized source. Analytically, for θ uniform over [0, 2π], the expected transmission E[cos²θ] = ½. A 1,000,000-sample Monte Carlo simulation returned 0.5001, confirming the result.

% Task 3: Power transmitted through polarizer for random polarization
N = 1e6;                          % number of samples
theta = 2*pi*rand(N,1);           % uniform random angles in [0, 2pi]
T = cos(theta).^2;                % Malus's law transmission
avg_transmission = mean(T);       % average power fraction
fprintf('Average transmission = %.4f (expected 0.5)\n', avg_transmission);

figure;
histogram(T, 50, 'Normalization', 'pdf');
xlabel('Transmission fraction cos^2(\theta)');
ylabel('Probability density');
title(sprintf('Random Polarization through Polarizer (avg = %.4f)', ...
      avg_transmission));
grid on;

Task 4 — Modulation and end-to-end link

Full binary link: 5 mW laser at λ = 1.55 µm → vertical polarizer → electro-optic modulator (V_π = 5 V) → quarter-wave plate → horizontal polarizer → multimode fiber → PIN photodiode (η_q = 80%).

At 50 Mbps the intermodal dispersion budget allows a maximum fiber length of ≈ 297.9 m (delay per unit length 33.56 ns/km, half-bit-period criterion). Photodiode responsivity worked out to 0.998 A/W.

View on GitHub

EECE 532 Advanced Semiconductor Lasers · Binghamton University · MATLAB

Numerical simulation of a directly modulated semiconductor laser using coupled rate equations for carrier and photon density. The device is biased just above threshold, pulsed to a higher drive current for 100 ps, then returned to bias — the same scenario a laser sees under direct modulation in a data link.

Rate equations

Carrier and photon densities are coupled through a gain term that includes compression:

dn/dt = I/(qV) − n/τ − Γ·G(n,s)·s
ds/dt = Γ·G(n,s)·s − s/τ_p + β·n/τ

G(n,s) = v_g · (dg/dn) · (n − n₀) / (1 + ε·s)

Device parameters

Results

Threshold came out to n_th ≈ 1.19 × 10¹⁸ cm⁻³ and I_th ≈ 22.9 mA. Steady state at 1.1·I_th was found by integrating to convergence over 8 ns rather than assuming carrier clamping, so the transient starts from a self-consistent bias point.

When the current steps up, carrier density overshoots n_th before stimulated emission can respond. Once photon density spikes it clamps the carriers, triggering a damped oscillatory exchange — and a second burst on the step back down. Relaxation oscillation frequency for this parameter set landed at ~5–15 GHz.

% Part (a): Threshold current
% Threshold condition: Gamma * vg * dgdn * (nth - n0) = 1/tau_p
delta_nth = 1 / (Gamma * vg * dgdn * tau_p);
nth  = n0 + delta_nth;
I_th = q * V * nth / tau;

% Transient: 1.1*Ith -> 4*Ith for 100 ps -> back to 1.1*Ith
I_func = @(t) I_bias + (I_high - I_bias) * (t >= t_on & t <= t_off);

opts   = odeset('RelTol', 1e-8, 'AbsTol', 1e-10, 'MaxStep', 0.2e-12);
t_eval = linspace(0, t_end, 8000);
[t_sol, y_sol] = ode45(@(t, y) rate_eqs(t, y, I_func, q, V, tau, tau_p, ...
    beta, Gamma, vg, dgdn, n0, epsilon), t_eval, y0, opts);

function dydt = rate_eqs(t, y, I_func, q, V, tau, tau_p, beta, ...
                         Gamma, vg, dgdn, n0, epsilon)
    n = max(y(1), 0);
    s = max(y(2), 0);
    I = I_func(t);
    G = vg * dgdn * (n - n0) / (1 + epsilon * s);
    dndt = I/(q*V) - n/tau - Gamma*G*s;
    dsdt = Gamma*G*s - s/tau_p + beta*n/tau;
    dydt = [dndt; dsdt];
end

View on GitHub

Bachelor's Thesis · SUNY Polytechnic Institute · Co-author (Processing)

Integrating Optical Technologies for Enhanced Processing, Memory, and Bandwidth Scalability in Next-Generation Computing. As Moore's Law approaches practical limits, this thesis examines how photonic technologies can move past the thermal, delay, and power-density constraints of silicon.

My section covered optical processing: photonic integrated circuits, all-optical computing architectures, and clock speeds beyond what electronic logic supports. Co-authors covered photonic memory and data-center scalability.

Technologies surveyed

Written with Owen Lawrence (memory) and Devin DiTucci (scalability / data centers) at SUNY Polytechnic Institute, Utica, NY.

View on GitHub

Personal project · KiCad · In progress

A regulated LM317 constant-current driver designed in KiCad that holds output current constant independent of supply voltage. The set resistor fixes the current through I = 1.25 / R_set — a 62 Ω resistor gives roughly 20 mA — with the regulator dissipating about 0.12 W from a 9 V supply, low enough to run without a heatsink.

The planned test is a supply sweep from 7 V to 12 V confirming output current stays flat, which is a line-regulation measurement. That same test carries over directly to the laser-diode version.

From there the board branches toward a protected laser-diode driver with current limiting and modulation — laser diodes fail fast under current transients, so the protection circuitry is the real engineering content.

EECE 501 Analog Circuit Design · Binghamton University · LTspice

An 8th-order band-pass filter built by cascading a 4th-order high-pass and a 4th-order low-pass biquad section, giving a 1 kHz – 50 kHz passband with more than 45 dB of stopband attenuation.

Stability was checked by solving each biquad's pole locations analytically and confirming all poles lie in the left half s-plane, then validating with a transient step response that decayed without oscillation.

The design was verified in LTspice with AC analysis from 1 Hz to 10 MHz and transient simulation — recovering a clean 30 kHz tone from a white-noise-corrupted input.

EECE 501 Analog Circuit Design · Binghamton University · LTspice

An NMOS cascode amplifier using an active current-source load in place of a resistive one. Swapping the load raises output resistance by orders of magnitude, and gain rises with it.

Transconductance and output resistance were hand-calculated from SPICE Level-1 model parameters (KP = 200 µA/V², V_T = 1.0 V, V_ov = 0.707 V) before simulating, then gain and bandwidth were verified against LTspice AC analysis from 10 Hz to 100 MHz.

Summer Undergraduate Research Program (SURP) · SUNY Polytechnic Institute · Python

An autonomous robot that follows a specific person while carrying items over long distances — aimed at elderly users and anyone needing mobility assistance. The build combines computer vision, embedded control, and power electronics.

Motor drive and power

I built the motor-drive subsystem around an L298N H-bridge running from a dual-battery supply, keeping motor power and logic power on separate rails so switching transients on the motor side don't disturb the Raspberry Pi. Speed and direction are set by Raspberry Pi PWM, with PID closed-loop control maintaining a safe following distance.

Vision and navigation

A YOLO-based vision system segments the camera frame into directional zones — forward, backward, left, right, stop — and the robot continuously reprocesses target coordinates to adjust movement in real time. Ultrasonic and infrared sensors handle collision avoidance, and a servo controls the storage compartment lid.

Hardware

Software

Supported by SURP 2024, with thanks to Dr. Gunyaz Ablay and Mcdhellyne Edwards. The repository includes the official SURP research poster.

View on GitHub Research poster (PDF)

Semiconductor Experimentation Research · SUNY Polytechnic Institute

A hands-on fabrication study of how silicon dioxide grows on silicon wafers, combining furnace processing, Deal–Grove modeling, and SEM characterization. SiO₂ is the gate oxide, insulator, and passivation layer that modern CMOS depends on, so controlling its thickness precisely is a core process-engineering problem.

Process

Modeling

Oxidation was simulated in nanoHUB Process Lab using the Deal–Grove linear-parabolic model across 850 °C, 975 °C, 1100 °C, and 1225 °C, studying oxide thickness against time and temperature. Arrhenius plots were used to estimate activation energy and characterize the reaction kinetics governing growth.

Characterization and findings

SEM imaging confirmed silicon and oxygen at the surface and the presence of the SiO₂ layer — and revealed nanowire structures that formed unexpectedly after oxidation. That discovery took second place at the 2024 SUNY Polytechnic Student Project Showcase and may warrant follow-up for nanoelectronics and sensing applications.

With Leo Bruno and Owen Lawrence. Faculty advisor: Dr. Iulian Gherasiou, SUNY Polytechnic Institute, College of Engineering.

View on GitHub Full paper (PDF)