The demand for datacenters computing power continues to grow, fueled by AI and cloud-computing applications. Traditional copper connections can no longer sustain the speed requirements and, as a result, optical networks have become the only viable solution for multi-Gbps intra-rack, intra-server and intra-cluster data transfers. In the context of optical links, the Transimpedance Amplifier (TIA) is the most critical block of the analog receiver chain, since it's the first amplification stage of the weak signal current generated by the receiving photodiode. This thesis explores multiple design strategies for PAM4 high-speed TIAs in 3nm FinFET technology and presents the best solutions for such an advanced node. This work begins with a system-level analysis of the optical receiver chain and explores how various standard TIA architectures perform in this technology, in terms of bandwidth, noise and power. Particular attention has been given to the characterization of the receiving photodiode and of the transmission line, which are major contributors to parasitic effects in Multi-GHz analog design. Then, the main limitation of the technology is identified: the gate-drain capacitances of transistors is not negligible as in planar ones and a folded-cascode topology is proposed, in order to address this particular design challenge. Two distinct variants of the folded-cascode topology are implemented, each of them provided with a replica bias structure and a Low Drop-Out regulator (LDO) for precise supply voltage regulation. The first configuration is a folded-cascode TIA, with input and output inductive peaking, while the second one has the same input peaking technique, but employs a T-coil at the output, in order to drive the capacitive load at the output node. For both configurations, key performance metrics, PVT corners and a comparison with existing topologies are presented. The thesis ends with one of the possible future developments: a differential implementation to further enhance signal-to-noise ratio.
La crescente richiesta di potenza di calcolo nei data center è spinta sopratutto dai recenti sviluppi riguardanti l'intelligenza artificiale e il cloud computing. Le connessioni in rame non sono piu' in grado di sostenere le velocità richieste e, di conseguenza, i collegamenti in fibra ottica tra rack, server e cluster, sono l'unica soluzione possibile per trasferimenti a centinaia di Gbps. Nel contesto dei collegamenti ottici, l'amplificatore a transimpedenza (TIA) rappresenta il blocco più critico della catena di ricezione del segnale, essendo il primo stadio di amplificazione della corrente generata dal fotodiodo ricevitore. Questa tesi esplora diverse architetture per amplificatori a transimpedenza ad alta velocità per segnali modulati in PAM4. Vengono poi presentate le soluzioni più adatte ad essere implementate in tecnologia FinFET 3 nm. La tesi inizia con un'analisi del sistema di ricezione ottico, valutando le prestazioni di banda, rumore e potenza di diverse architetture di TIA. Particolare attenzione viene dedicata alla caratterizzazione del fotodiodo e della linea di trasmissione, che costituiscono le principali fonti di parassiti in design a multi-GHz. Viene poi identificata la principale limitazione tecnologica all'interno della soluzione circuitale: le capacità gate-drain dei FinFET non sono più trascurabili come in precedenti tecnologie planari. Per affrontare questo problema, viene proposta un'architettura di tipo folded-cascode. Sono state implementate due varianti della topologia folded-cascode, entrambe dotate di replica bias e di un regolatore LDO per garantire una corretta tensione di alimentazione. La prima configurazione propone un TIA basato su amplificatore folded-cascode con peaking induttivo sia in ingresso che in uscita, mentre la seconda impiega la stessa tecnica di peaking in ingresso, ma con T-coil in uscita, in modo da pilotare il carico capacitivo. Per entrambe le configurazioni vengono presentati i principali parametri prestazionali, analisi PVT e un confronto con topologie esistenti. La tesi si conclude con una possibile evoluzione futura: l'implementazione differenziale finalizzata a migliorare ulteriormente il rapporto segnale-rumore.
Design of high performance Transimpedance Amplifiers for the next generation optical links in 3nm FinFET
Garbi, Alessandro
2024/2025
Abstract
The demand for datacenters computing power continues to grow, fueled by AI and cloud-computing applications. Traditional copper connections can no longer sustain the speed requirements and, as a result, optical networks have become the only viable solution for multi-Gbps intra-rack, intra-server and intra-cluster data transfers. In the context of optical links, the Transimpedance Amplifier (TIA) is the most critical block of the analog receiver chain, since it's the first amplification stage of the weak signal current generated by the receiving photodiode. This thesis explores multiple design strategies for PAM4 high-speed TIAs in 3nm FinFET technology and presents the best solutions for such an advanced node. This work begins with a system-level analysis of the optical receiver chain and explores how various standard TIA architectures perform in this technology, in terms of bandwidth, noise and power. Particular attention has been given to the characterization of the receiving photodiode and of the transmission line, which are major contributors to parasitic effects in Multi-GHz analog design. Then, the main limitation of the technology is identified: the gate-drain capacitances of transistors is not negligible as in planar ones and a folded-cascode topology is proposed, in order to address this particular design challenge. Two distinct variants of the folded-cascode topology are implemented, each of them provided with a replica bias structure and a Low Drop-Out regulator (LDO) for precise supply voltage regulation. The first configuration is a folded-cascode TIA, with input and output inductive peaking, while the second one has the same input peaking technique, but employs a T-coil at the output, in order to drive the capacitive load at the output node. For both configurations, key performance metrics, PVT corners and a comparison with existing topologies are presented. The thesis ends with one of the possible future developments: a differential implementation to further enhance signal-to-noise ratio.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_10_Garbi_Tesi_01.pdf
non accessibile
Descrizione: Testo tesi
Dimensione
8.63 MB
Formato
Adobe PDF
|
8.63 MB | Adobe PDF | Visualizza/Apri |
|
2025_10_Garbi_Executive_Summary_02.pdf
non accessibile
Descrizione: Testo executive summary
Dimensione
1.94 MB
Formato
Adobe PDF
|
1.94 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/10589/243878