張耀仁: 6月 2025

星期一, 6月 30, 2025

ＡＩ在PCB產業的應用

1. Traveling Salesman Problem 簡稱TSP在工業界有重要應用，包括物流（UPS/Amazon配送路線優化）、製造業（電路板鑽孔、機器人組裝路徑）、電信（網路路由、線路安裝）和能源（電網維護、管線檢查）。凡是需要造訪多個地點同時最小化成本、時間或距離的場合都適用。現代變體能處理容量限制、時間窗口等實際約束。企業使用OR-Tools、Gurobi等專業軟體解決這些問題，透過優化倉儲揀貨、切割模式、3D列印路徑和車隊管理等作業，往往能節省數百萬成本。

視覺化 Traveling Salesman Problem 使其可以改變網路節點個數，或是其他你希望的方式

2. 給定一個電路板，進行鑽孔程序模擬的優化 (建議 AI 一個演算法，例如A*, Greedy Method...)

電路板鑽孔程序模擬

星期三, 6月 11, 2025

Syllabus 科技英文與科普傳播 2025 秋天

上課方式: 使用科技英文學習掌握AI運用技巧，從實做中學習AI領域的技術應用能力與未來科學工程素養。

講授，工作坊，作業

學生請自行攜帶手機，平板或筆電。

本課程特色

9/10/2025 - Week 1: Introduction

百工如何使用ＡＩ (SVG)

受影響職業

最好的程式語言 (Python? JavaScript?)

課程主題一覽表 svg

9/17/2025 - Week 2: Boosting human brains with AI

增強ＡＩ的能力

9/24/2025 - Week 3: Reasoning with AI for Complex Constrained Problems

推理模式 special topic: Smart Dog Ball Retrieval

極紫外光（EUV）光源, IEEE Spectrum: The Tiny Star Explosions Powering Moore’s Law

AI半導體廠製程優化

10/1/2025 - Week 4

Timeline for dinner scheduling

Generate programs for dinner schedule (AI半導體廠製程優化)

River Puzzles (半導體晶圓廠情境 Part 1)

11/5 - 期中考不上課

11/12/2025 - Week 9: Visualization

Update of new Claude Sonnet 4.5

AI 互動的基礎：如何與 AI 有效溝通 (Prompt 基礎篇) Pt 1

工程動畫 Pt 1，精選

11/19/2025 - Week 10

Catch Up

工程動畫 Pt 2

AI 互動的基礎：如何與 AI 有效溝通 (Prompt 基礎篇) Pt 2

11/26/2025 Scientific Animation

Blue LED

3d 物件與互動 (水與冰)

Cat coat colors: 貓的花色遺傳模擬

AI for Scientific Discovery

12/3/2025 - Week 12 High Complexity Animation

Follow up #11

隨機程序模擬

Enhance visualization (Mul. Iter. Merg.)

12/10/2025 - Week 13: AI for Optimization

演算法視覺化

Optimize Circuit Design

12/17/2025 - Week 14: Reinforcement Learning

follow up (Circuit Optimization)

Reinforcement Learning

聖誕節放假

12/31/2025 - Special Topics

Fractals: 美麗碎形世界

Build an AI APP

AI 財務大數據分析

Capstone

Scientific Visualization 2

Backup

真實世界 3d 模型製作 from Pictures

Special Topic

Games

Further Materials

Taiwan Determined to Remain ‘Indispensable’ Source of Best Chips (Bloomberg)
BBC Nature
Japan top inventions | NHK
Science view | NHK
Science that can save your life | NHK

星期二, 6月 10, 2025

Implementation of systems based on paper/report

ATM Simulator (paper, download)

shower training (paper, download)

revision batch 1

revision batch 2

revision batch 3

Agility Training (artifact, 2d version), paper

Agility Training (share, 3d version), paper

hand washing (paper, download)

Community-based Purchasing skills, paper

照護app (from paper to app)

VLM-based Navigation (From paper to app)

3d Safety assessment

artifact (first person perspective)

artifact (Fractal tree versions)

artifact (first person perspective)

artifact (fractal tree version)

artifact (drone view)

artifact (first person perspective)

artifact (non Fractal version)

artifact (fractal version)

share (artifact)

share (source)

artifact

Strategic thinking

Devise a strategy for the dog in the picture to retrieve its ball that fell into the pond without getting its body wet, with no humans nearby to help."

This is a problem-solving or creative thinking exercise asking someone to come up with a plan for how a dog could get its ball back from water without getting wet and without human assistance.

Animate your strategy in svg or in 3d.

Solution 1

3D Use of float bridge ,

3D use of float bridge (thought out by Claude itself) Veo 2 (video w/ base image)

float bridge Imagen 4, another, Veo 2 (video)

Solution 2

float + paddling inspired by student

Solution 3

3D, base 2D float push, wave making, another 2D

solution 4

Noodle Nudge Veo2 imagen 4

many clever strategies by Gemini 2.5 Pro

original video on social media

星期一, 6月 02, 2025

EX# Scientific Visualization

1. 科學現象3d模擬(雷射，超導體，超流體，中子星碰撞...)，選取以下一個題目，加上自選一個主題

a. 科學模擬 3d 化(使用 three.js 模擬 3d )

Black hole-star binary system simulation

黑洞與恆星雙星系統

黑洞撞擊

b. 使用 three.js 模擬 3d

2. 雷射(2d 模擬 energy pumping)

雷射(3d 模擬) 使用3d量子場

CO2 Sequestration Process, in 3D simulation. 二氧化碳封存地底岩層過程

Algorithms

ＱuickSort 是常用的數字排序演算法，我們使用ＡＩ來視覺化它的排序過程，可以清楚看到數字如何逐漸由小排到大。

QuickSort 演算過程視覺化。

Illustrate in SVG the code.

汽車導航常常需要計算兩地之間最短路徑。我們用ＡＩ視覺化一個有名的最短路徑演算法（Dijkstra's Algorithm），你可輸入網路節點數，或是你希望的視覺化方式。

Shortest Path Algorithm

Traveling Salesman Problem 簡稱TSP

(一個推銷員要拜訪所有客戶城市，每個城市只能拜訪一次，最後要回到出發城市，請為他/她計算最短的拜訪路徑)

TSP在工業界有重要應用，包括物流（UPS/Amazon配送路線優化）、製造業（電路板鑽孔、機器人組裝路徑）、電信（網路路由、線路安裝）和能源（電網維護、管線檢查）。凡是需要造訪多個地點同時最小化成本、時間或距離的場合都適用。現代變體能處理容量限制、時間窗口等實際約束。企業使用OR-Tools、Gurobi等專業軟體解決這些問題，透過優化倉儲揀貨、切割模式、3D列印路徑和車隊管理等作業，往往能節省數百萬成本。

(樸素) Visualize TSP (Traveling Salesman Problem) by A* search 使其可以改變網路節點個數

(美學) 視覺化 A* for Traveling Salesman Problem 使其可以改變網路節點個數

給定一個電路板，進行鑽孔程序模擬的優化 (建議 AI 一個演算法，例如A*, Greedy Method...)

A* Drilling

電路板鑽孔程序模擬

AI 在 PCB 產業的應用

A* 演算法

以下迷宮問題，比較Ａ* search 與 Dijkstra 求解過程

A* search vs. Dijkstra for a maze

Tower of Hanoi

Challenge

3+. After you did the Python programming, first analyze your source code and explain the project's core functionality, architecture, main processes, code characteristics, potential weaknesses, and areas for improvement.

Amplifier Gain and Frequency Response Using AI

電子學問題ＡＩ求解方式

standard strategy

AI for solving amplifier electronics (課堂講義內有參考解答)

solve for gain

do not ignore bias current

calculate miller effect

calculate poles for lower cut off

calculate poles for higher cut off

plot bode plot gain vs. Hz

plot bode plot phase vs. Hz

Solve for the gain. Plot frequency response for the following amplifier circuit.

Source

artifact

AI Model Results:

Grok 4: Share Link (considers parasitic capacitance Cπ ≈ 50 pF, Cμ ≈ 2 pF) - Bode Plot (188 kHz)
Claude: Bode Plot gives 188 kHz,
ChatGPT 5: Share Link gives 187 kHz
Gemini 2.5 Pro: Share Link gives 188 kHz

Note: Lower cut-off comes from coupling capacitors (not shown here).

Generate professional grade Bode plot.

Using AI tools to generate professional plot using Matplotlib

Bode plot

ChatGPT integrates Python environment so it directs generates Bode plot.

Bode plot

React (Claude) - Python w/Matplotlib - - Colab - Plot

Colab

Bode plot

1. React (Claude) - AIStudio - Python w/ Matplotlib - Colab - Plot

2. Reasoning (AIStudio) - AIStudio - Python w/ Matplotlib - Colab - Plot

我用Claude計算電晶體放大器，得出頻率響應理論值 2.4 MHz。然後我給他一組Grok 算出來的數據是 125 kHz，兩者差了20倍。

我沒有跟Claude 說這組數據哪裡來的，他稱讚我的數據是應該是用Spice 專業軟體跑出來的，而且提醒我他的計算是根據純理論，因為Spice 可以考慮更多細節去電路模擬，所以Claude認為我的數據比較值得參考。

接下來我跟他講這組數據是 Grok 4 算的，Claude 馬上變換語氣，說頻率響應不可能是125 kHz，業界標準至少在1 MHz 以上，說Grok:

Does not make any sense (完全沒有道理）
Against common sense (違背常識）
Distrust (不值得信賴）

可是剛剛不是還在稱讚「我的模擬」做得好？以前Claude 還會稱讚別家的AI有什麼優勢，現在看起來不演了。

Amplifier Circuit Analysis

Solve for the gain, plot freq response.

Share (Source)

share (Claude running Python internally on VM)

Hybrid-π Model, With bypass capacitors, without ignoring base current loading effects on bias networks, Av=-165 (Share including frequency response).
-194 for Vc vs. Vbase, Ibase ignored (Share)

Gemini 2.5 Pro gain & lower cut off freq match Claude, 4kHz.

Solve for the gain. Plot frequency response for the following amplifier circuit.

Source

share (Av=-2.31, Lower Cutoff 20 Hz, Upper Cutoff 10.46MHz)

Analysis with Cπ = 22.1 pF, Cμ = 3.0 pF

20Hz is assumed, not by calculation

Analysis with Cπ = 22.1 pF, Cμ = 3.0 pF:

Bode plot, fH = 10.46 MHz, RE not bypassed

Two-Stage Amplifier Analysis

Solve for the gain, plot frequency response.

Source

Schematic

colab unoptimized (share)

colab optimized (share)

colab increasing C2 from 1 to 10μF (other parameters differ from above)(share)

bypass 47, coupling 10 micro F

(gain approximate)

Artifact, Share (Cπ2=15 pF, Cμ2=3pF). Result: freq response, gain=8513, low cut off=46.5 Hz, high cut off=312kHz
Gemini: Result gain=8509, low cut off=46.5 Hz, high cut off=296 kHz @Cπ2=15 pF, Cμ2=3pF (good match!)

Compare

Artifact, Share. Result: gain=8559 (Hybrid-π Model, With bypass capacitors, without ignoring base current loading effects)
Artifact. Result: gain=9460 (Ignored base current loading effects)
comprehensive comparison simplified, do not ignore bias network, do not ignore base current

Can we ask for Bode plot without finding lower/higher cut-off freq?