Developments in the information technology field have always been strictly linked to the evolution of the associated electronics circuits and systems, but this trend, along with Moore’s law, is nearing its end. Applications such as Artificial Intelligence (AI) and data centers are growing in both dimension and complexity, pushing electronics to its physical limits. Increasing the frequency and size of a digital electronic circuit results in a more than proportional increase of power consumption, following the relation P ≈ f CV^2. Moreover, copper wires are gradually being substituted with faster optical data lines in many applications where data integrity is mandatory. For these reasons, an alternative to standard electronic systems must be found and a cooperation of photonics and digital electronics seems to be a step in the right direction. Photonics, by means of interferometer meshes, can mitigate all of these problems. Data transmission, at least for the photonic domain path, has a negligible power loss, independent on the data rate, giving us the possibility to operate at much higher frequencies and for longer distances without the risk of losing the signal integrity because all the transmission is done at the speed of light. Moreover, thanks to the linear behavior of most optical components, it is possible to design photonic structures that perform mathematical operations, namely Matrix-Vector Multiplication (MVM) which is the fundamental operation performed in neural networks training. However the photonic alternative does not come without challenges. Process non-idealities and environmental contributions strongly affect optics; thermal drifts or crosstalk change the behavior of optical components, resulting in an unreliable and unexpected result. The solution to these problems comes from the cooperation of both worlds. An optical mesh is used in combination with an electronic control layer in order to achieve, thanks to a specifically designed control algorithm, a precise and stable behavior. To achieve this, a FPGA-based control loop is used in our setup, allowing reprogrammability and flexibility for the control system to be used with larger mesh sizes as well. The sensing is done by means of transparent photodiodes, characterized by minimal insertion losses, needed to measure power and phase at any point of the mesh without affecting the signal light, while the actuation is done by changing locally the optical fiber temperature with TiN heaters. The focus of this thesis was to validate and optimize the control chain for a 3x3 optical mesh realized with Mach Zender interferometers. The whole system was then characterized and validated using a 1550nm laser, confirming the accuracy reached with the previous 2x2 mesh, with a setting error lower than 2% in some cases, and demonstrating the possibility to extend this control to larger meshes paving the way for electro-optical computing.
Lo sviluppo della tecnologia dell’informazione è sempre stato strettamente legato all’evoluzione dell’elettronica associata, ma questa tendenza, insieme alla legge di Moore, sta giungendo al termine. Applicazioni come l’intelligenza artificiale e i data center stanno crescendo sia in dimensioni che in complessità, spingendo l’elettronica ai suoi limiti fisici. L’aumento della frequenza e delle dimensioni di un circuito elettronico digitale comporta un incremento più che proporzionale del consumo di potenza, secondo la relazione P ≈ f CV^2 ; inoltre, esistono molte applicazioni in cui le linee dati ottiche sono indispensabili per la salvaguardia dell’integrità dei dati. Per questa ragione, è necessario trovare un’alternativa, e una cooperazione tra la fotonica in silicio e l’elettronica digitale sembra essere un passo nella giusta direzione. La fotonica, per mezzo di reti di interferometri, può risolvere o evitare quasi tutti questi problemi. La trasmissione dei dati, almeno per il percorso nel dominio ottico, ha una perdita di potenza trascurabile e indipendente dalla velocità di trasmissione, dandoci la possibilità di operare a frequenze molto più elevate e su distanze maggiori senza il rischio di perdere l’integrità del segnale, poiché tutta la trasmissione avviene alla velocità della luce. Tuttavia, l’alternativa fotonica non è priva di sfide. Le non-idealità di processo e i contributi ambientali influenzano l’ottica allo stesso modo dell’elettronica, o anche di più; le derive termiche o il crosstalk modificano il comportamento rispetto a quello previsto, portando a un risultato inaffidabile e inaspettato. La soluzione a questi problemi deriva dalla cooperazione di entrambi i mondi. Una rete ottica viene utilizzata in combinazione con una catena di controllo elettronica per ottenere, grazie a un algoritmo di controllo appositamente progettato, un comportamento preciso e stabile. Per raggiungere questo obiettivo, nel nostro setup viene utilizzato un anello di controllo basato su FPGA, che consente riprogrammabilità e flessibilità affinché il sistema di controllo possa essere utilizzato anche con reti di dimensioni maggiori. Il rilevamento (sensing) viene effettuato tramite fotodiodi trasparenti, caratterizzati da perdite di inserzione minime, necessari per misurare potenza e fase in qualsiasi punto della rete senza influenzare il segnale luminoso, mentre l’attuazione avviene modificando localmente la temperatura della fibra ottica con riscaldatori in TiN (nitruro di titanio). L’obiettivo di questa tesi è stato validare e ottimizzare la catena di controllo per una rete ottica 3x3 realizzata con interferometri di Mach-Zehnder. L’intero sistema è stato quindi caratterizzato e validato utilizzando un laser a 1550 nm, confermando l’accuratezza raggiunta con la precedente rete 2x2, con un errore di settaggio in alcuni casi inferiore al 2%, e dimostrando la possibilità di estendere questo controllo a reti di dimensioni maggiori, aprendo così la strada al calcolo elettro-ottico.
High precision control optimization and validation of n-order Mach-Zehnder meshes for complex-values optical computing
MAURI, EDOARDO
2024/2025
Abstract
Developments in the information technology field have always been strictly linked to the evolution of the associated electronics circuits and systems, but this trend, along with Moore’s law, is nearing its end. Applications such as Artificial Intelligence (AI) and data centers are growing in both dimension and complexity, pushing electronics to its physical limits. Increasing the frequency and size of a digital electronic circuit results in a more than proportional increase of power consumption, following the relation P ≈ f CV^2. Moreover, copper wires are gradually being substituted with faster optical data lines in many applications where data integrity is mandatory. For these reasons, an alternative to standard electronic systems must be found and a cooperation of photonics and digital electronics seems to be a step in the right direction. Photonics, by means of interferometer meshes, can mitigate all of these problems. Data transmission, at least for the photonic domain path, has a negligible power loss, independent on the data rate, giving us the possibility to operate at much higher frequencies and for longer distances without the risk of losing the signal integrity because all the transmission is done at the speed of light. Moreover, thanks to the linear behavior of most optical components, it is possible to design photonic structures that perform mathematical operations, namely Matrix-Vector Multiplication (MVM) which is the fundamental operation performed in neural networks training. However the photonic alternative does not come without challenges. Process non-idealities and environmental contributions strongly affect optics; thermal drifts or crosstalk change the behavior of optical components, resulting in an unreliable and unexpected result. The solution to these problems comes from the cooperation of both worlds. An optical mesh is used in combination with an electronic control layer in order to achieve, thanks to a specifically designed control algorithm, a precise and stable behavior. To achieve this, a FPGA-based control loop is used in our setup, allowing reprogrammability and flexibility for the control system to be used with larger mesh sizes as well. The sensing is done by means of transparent photodiodes, characterized by minimal insertion losses, needed to measure power and phase at any point of the mesh without affecting the signal light, while the actuation is done by changing locally the optical fiber temperature with TiN heaters. The focus of this thesis was to validate and optimize the control chain for a 3x3 optical mesh realized with Mach Zender interferometers. The whole system was then characterized and validated using a 1550nm laser, confirming the accuracy reached with the previous 2x2 mesh, with a setting error lower than 2% in some cases, and demonstrating the possibility to extend this control to larger meshes paving the way for electro-optical computing.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_03_Mauri_Executive Summary.pdf
accessibile in internet per tutti
Descrizione: testo executive summary
Dimensione
2.95 MB
Formato
Adobe PDF
|
2.95 MB | Adobe PDF | Visualizza/Apri |
|
2026_03_Mauri_Tesi.pdf
accessibile in internet per tutti
Descrizione: testo tesi
Dimensione
86.82 MB
Formato
Adobe PDF
|
86.82 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/251242