傳統的光譜儀笨重又不利攜帶，且通常需藉由脆弱的光纖當作媒介來接收光源，以上特點都使傳統光譜儀在使用上有諸多限制。近年來因為環保意識與食安問題等觀念抬頭，光譜儀的應用變得更加多元且重要，因此新型的可攜式微型光譜儀便成為眾人積極發展的目標。
本實驗便是藉由濾波器陣列、光檢測器陣列、以及讀出電路(Readout-IC)彼此搭配而組成一個光譜儀模組，藉由微影製程可將濾波器與光檢測器做到微米尺寸，大幅縮小元件的體積，且透過晶圓製造的技術使元件量產化、降低製造成本。但目前讀出電路晶片尚未開發完成，故整個模組暫時建立在電路板上，但整體架構與讀出晶片是一樣的。
本實驗能重建之光譜範圍為900nm~1700nm，在搭配電路板情況下，解析度約為12.5nm，將來讀出晶片完成後，解析度更可期待提高到0.8nm。本實驗中我們已經成功重建紅外LED的光譜，對於單顆LED重建光譜的最小相對誤差為0.045624、平均相對誤差為0.083593;對於多顆LED重建光譜的最小相對誤差為0.095399、平均誤差為0.114901。

In recent years, the applications of spectrometer are more diverse due to both the rise of environmental awareness and food safety scandal. As a result, the portability as well as convenience become the essential part for spectrometer. However, the conventional spectrometer is usually heavy and enormous, which makes it unable to take along everywhere. In addition, the frail optical fiber also causes a problem when we want to measure sample freely.
In this work, we combine filter array(FA), photodetector array(PDA) and readout circuit(ROIC) to form a miniature spectrometer module. However, because the ROIC is still under testing, we demonstrate the same model by FA and several PD on PCB. Besides, in order to get the object spectrum, output signals have to be converted by solving the simultaneous equations.
We demonstrate a prototype miniature spectrometer which can sense from 900nm to 1700nm with resolution of 12.5nm. In the future, the resolution can be improved to 0.8nm when the ROIC is finished. In addition, we have successfully reconstructed the spectra of NIR-LED and the minimum relative error is 0.045624 for single-LED and 0.095399 for multi-LED. Besides, the average relative error is 0.083593 for single-LED and 0.114901 for multi-LED.

口試委員會審定書 #
中文摘要……………………………………………………………………i
ABSTRACT ii
誌 謝………………………………………………………………………iii
CONTENTS iv
LIST OF FIGURES vii
LIST OF TABLE xii
Chapter 1 Introduction 1
1.1 Origin and Development 1
1.2 Motivation and Purpose 5
Chapter 2 The Basis of Theory 10
2.1 Rebuilding Spectrum by Solving Simultaneous Equations 10
2.2 Ill-posed System and Condition Number 11
2.3 Regularization Techniques and optimizing regularization parameter methods 24
2.3.1 Singular Value Decomposition (SVD) 25
2.3.2 Tikhonov Regularization 27
2.3.3 Truncated SVD (TSVD) 29
2.3.4 Least Squares with a Quadratic Constraint(LSQR) 30
2.3.5 L-curve 31
2.3.6 Generalized cross-validation (GCV) 32
Chapter 3 Design and Fabrication 34
3.1 Model of the miniature spectrometer 35
3.2 Design of the Mask 37
3.3 Simulation of the Optical Propertie 40
3.4 Process Flow 42
3.5 Instruments for measuremen 48
3.5.1 Micro-point optical measurement system 48
3.5.2 Monochromator system, 34401A and SC-5000 49
Chapter 4 Results and Discussion 52
4.1 Using 5x5 Discrete-FA to Rebuild Monochromatic light 52
4.2 Using 8x8 Discrete_FA to Rebuild Single NIR-LED 55
4.2.1 Rebuilding 960nm-LED by Regularization 58
4.2.2 Rebuilding 1070nm-LED by Regularization 60
4.2.3 Rebuilding 1200nm-LED by Regularization 62
4.2.4 Rebuilding 1600nm-LED by Regularization 64
4.3 Using 8x8 Discrete_FA to Rebuild multiple NIR-LED 66
4.3.1 Rebuilding 960nm+1200nm-LED by Regularization 67
4.3.2 Rebuilding 1200nm+1600nm-LED by Regularization 69
4.3.3 Rebuilding 960nm+1600nm-LED by Regularization 71
4.3.4 Rebuilding 1070nm+1200nm-LED by Regularization 73
4.3.5 Rebuilding 960nm+1070nm-LED by Regularization 76
4.3.6 Rebuilding 1070nm+1600nm-LED by Regularization 78
Chapter 5 Conclusions 82
Chapter 6 Future Work …………...………………………………......83
REFERENCES 84

[1] R. C. Aster, B. Borchers, and C. H. Thurber, “Parameter Estimation and Inverse Problems,” New York: Elsevier Academic, 2005.
[2] P. C. Hansen, “REGULARIZATION TOOLS: A Matlab package for
analysis and solution of discrete ill-posed problems,” Numer. Algor.,
vol. 6, no. 1, pp. 1–35, Mar. 1994.
[3] C.-C. Chang, and H.-Y. Lin, “Spectrum reconstruction for on-chip spectrum sensor array using a novel blind nonuniformity correction method,” IEEE Sensors J., vol. 12, no. 8, pp. 2586–2593, AUGST. 2012.
[4] L. Elden, “Algorithms for regularization of ill-conditioned least-squares problems,” BIT Numerical Mathematics, vol. 17, no. 2, pp.134-145, 1977.
[5] P. C. Hansen, “The truncated SVD as a method for regularization,” BIT Numerical Mathematics, vol. 17, pp.543-553, 1986.
[6] https://ccjou.wordpress.com/ 線代啟示錄
[7] P. C. Hansen, T. K. Jensen & G. Rodriguez, “An adaptive pruning algorithm for the discrete L-curve criterion,” J. Comp. Appl. Math. Vol. 198, Issue 2, pp.483-492, 2007.