Engineering Workbench | Raihan Fauzy Ferdyansyah

IC / Mixed-Signal

Circuit design and process-temperature characterization.

IC design · Chipathon 2026 CMOS Schmitt Trigger GF180MCU · 3.3 V · 8T CMOS · Xschem / ngspice Schematic design and process-temperature characterizationNominal ΔVH 0.731 V · minimum ΔVH 0.712 V · 5 corners × 3 temperatures at 3.3 V Open record ↓Close record ↑

Topology

8T Schmitt trigger with regenerative core and output buffer

Designed in GF180MCU as a signal-conditioning block for the DVL receiver chain. The regenerative core creates hysteresis, while the output buffer restores non-inverting polarity and isolates the internal node.

PDK: GF180MCU
Supply: 3.3 V
Core: 8T CMOS

Xschem schematic of the GF180MCU 8T CMOS Schmitt trigger — **Xschem schematic** 6T regenerative core with M7/M8 output inverter buffer.

Hysteresis and process-temperature sweep summary for the GF180MCU Schmitt trigger — **Hysteresis / process-temperature summary** Nominal: VSPH 1.913 V · VSPL 1.181 V · ΔVH 0.731 V · center 1.547 V.

ngspice time-domain input and output waveform of the Schmitt trigger — **Time-domain response** Triangular input sweep and rail-to-rail output switching from the ngspice testbench.

Process-temperature coverage5 corners × 3 temperatures

Minimum hysteresis0.712 V

Center range1.475–1.615 V

GF180MCU/ngspice sweep summary across 15 process and temperature combinations.

Applied Data / ML

Held-out model evaluation and risk visualization.

Applied data / ML Early Warning Karhutla Riau · 5 km grid · 7-day hotspot risk · LightGBM Per-cell classification + interactive dashboard demoSemifinalist · University Track DATATHON 2026 Open record ↓Close record ↑

Contribution

Model evaluation + dashboard development

The task was to classify hotspot risk per cell over a seven-day window. I worked on model experimentation and evaluation and helped build the dashboard in this team project. Separating environmental features from hotspot history made the operational dependence explicit.

Open dashboard demo ↗

Grid: 5 × 5 km
Input window: 14 days
Target window: 7 days
Evaluation unit: Per cell

The study separates environmental inputs from an operational feature regime that includes hotspot history. Each cell receives one risk value for the full seven-day window.

LightGBM wildfire risk map for Riau on 25 September 2023 — **25 Sep 2023 · dry season** Source-paper risk-map illustration.

LightGBM wildfire risk map for Riau on 15 January 2023 — **15 Jan 2023 · wet season** Source-paper risk-map illustration.

Team data and modeling pipeline for the Riau wildfire forecasting project — **Team pipeline design** Multimodal inputs, 5 × 5 km grid standardization, 14-day input window, 7-day target, model comparison, and evaluation. My contribution focused on the modeling and dashboard layers.

Scope boundary. The study evaluates per-cell risk classification from 14 input days for the following seven-day window, t+1 through t+7. Risk maps are source-paper illustrations. Their embedded annotations are not independently verified final-run metrics. The linked demo currently labels its data as simulated. This work does not claim real-time fire detection or a real-time operational forecasting service.

Industrial / Embedded Systems

Industrial monitoring and embedded prototype work.

Industrial monitoringIndustrial Fire-Suppression MonitoringEbyte ME31 I/O · Modbus TCP/RTU · Node-RED · MySQL · TelegramMonitoring workflow + I/O and communication checksModbus read · state filtering · MySQL logging · dashboard · Telegram alerts Open record ↓Close record ↑

Project scope

Fire-suppression status monitoring workflow

The monitoring workflow needed to carry device status from Modbus I/O to a dashboard, event log, and alerts. I configured the I/O and built Node-RED state handling with MySQL logging and Telegram notifications. The evidence below follows those communication and state-processing paths.

Field I/O: Ebyte ME31
Protocols: Modbus TCP / RTU
Monitoring: Node-RED + MySQL

Ebyte ME31 I/O modules wired with LAN and 12 V supply during monitoring-system setup — **01 · Field / I/O**Module wiring, network addressing, and channel-level functional checks.

Node-RED flow for Modbus fire-suppression monitoring, state filtering, logging, alarms, and notifications — **02 · System logic**Modbus read → parsing / RBE → state handling → database, alarm, and notification paths.

Node-RED dashboard for fire-suppression status monitoring — **03 · Operator output**Fire-suppression status dashboard.

Telegram channel receiving fire-suppression monitoring status notifications — **Notification path**Status messages delivered through Telegram.

Scope boundary. This internship work focused on monitoring, Modbus communication, dashboard/database logic, notification flow, and device checks. PLC operation, direct sensor installation, and full plant-production integration were outside this internship scope.

Embedded / IoT / applied MLEMMA Smart CompostingSTM32 prototype · environmental sensing · actuation · IoT monitoring · Decision TreeTeam capstone · Best Impact · DGIP software copyrightSystem prototyping · offline Decision Tree with entropy and information gain Open record ↓Close record ↑

System context

Smart-composting monitoring and control prototype

The team prototype combines STM32 sensing, actuation, and IoT monitoring for composting. My contribution was system prototyping and a Python Decision Tree using entropy and information gain. The code establishes an offline classification proof of concept, with sensor-to-model mapping still incomplete.

Embedded: STM32
ML logic: Decision Tree
Recognition: Best Impact

EMMA smart-composting system architecture connecting STM32, sensors, actuators, monitoring, database, and Decision Tree — **System architecture**Embedded sensing / actuation connected to monitoring, storage, and decision logic.

Python implementation of entropy, dataset splitting, and information-gain search for EMMA Decision Tree — **Entropy / information gain**Entropy, dataset splitting, and information-gain search implemented directly in Python.

Python Node class and recursive Decision Tree construction used in the EMMA capstone — **Recursive tree construction**Node representation and recursive build logic.

EMMA Smart Composting team with the physical prototype and Best Impact recognition — **Prototype + recognition**Best Impact · Capstone DTETI UGM 2025. Software copyright registered with DGIP No. 001397406.

ML scope. The Decision Tree is presented as an offline proof of concept. The documented model inputs were not fully mapped to the prototype sensor set, and no final accuracy or F1 result is claimed here.

Supporting Hardware

Shorter build records from schematic to bench.

Discrete analog hardware 4-Stage BJT Amplifier 2N5550 · LTspice · KiCad · fabricated PCB Hand analysis → simulation → PCB → bench measurement 96.8 V/V measured gain · ≈100 V/V target Open record ↓Close record ↑

Target ≈100 V/V. Measured gain was 96.8 V/V after hand analysis, LTspice simulation, PCB implementation, and oscilloscope validation.

LTspice schematic of the four-stage 2N5550 BJT amplifier — **Simulation**LTspice circuit schematic.

Fabricated and assembled four-stage BJT amplifier PCB — **Hardware**Fabricated and assembled board.

Oscilloscope bench measurement for the four-stage BJT amplifier — **Bench**Oscilloscope measurement of voltage gain.

Analog / sensor hardware IR Proximity Sensor IR LED · phototransistor · BC547 · LM3914 · PCB Simulation → PCB → physical prototype → bench test Open record ↓Close record ↑

IR source, phototransistor detector, analog stage, and LM3914 indication. Simulated, laid out, assembled, and bench tested.

Fabricated proximity sensor board — **Implementation**Physical PCB prototype.

Bench measurement during proximity sensor distance testing — **Bench test**Distance-response measurement.