A Process and Temperature Insensitive CMOS Linear TIA for 100 Gb/s/$\lambda$  PAM-4 Optical Links

Lakshmikumar, K.R.; Kurylak, Alexander; Nagaraju, Manohar; Booth, Richard; Nandwana, Romesh Kumar; Pampanin, Joe; Boccuzzi, V.

doi:10.1109/jssc.2019.2939652

Cited by 47 publications

(18 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Using this expression, we can calculate the minimum detected power necessary to have a SNR = 10 dB, which is a conservative number. Using commercially available photodetectors [59] with = 0.85 A/W, I d = 30 nA, B = 10 GHz, R L = 50 Ω and reported TIAs [60] with input referred noise density of 16.7 pA/ÝHz we obtain that the minimum optical power that has to be photodetected is −21.6 dBm, or 6.96 * 10 -3 mW. Using these values, we can estimate the number of layers that can be concatenated.…”

Section: Appendix B Scalabilitymentioning

confidence: 99%

Reconfigurable Activation Functions in Integrated Optical Neural Networks

Campo¹,

Pérez²

2022

IEEE J. Select. Topics Quantum Electron.

View full text Add to dashboard Cite

The implementation of nonlinear activation functions is one of the key challenges that optical neural networks face. To the date, different approaches have been proposed, including switching to digital implementations, electro-optical or all optical. In this article, we compare the response of different electro-optic architectures where part of the input optical signal is converted into the electrical domain and used to self-phase modulate the intensity of the remaining optical signal. These architectures are made up of Mach Zehnder Interferometers (MZI) and microring resonators (MRR). We have compared the corresponding transfer functions with commonly used activation functions in state-of-theart machine learning models and carried out an in-depth analysis of the capabilities of those architectures to generate the proposed activation functions. We demonstrate that a ring assisted MZI and a two-ring assisted MZI present the highest expressivity among the proposed structures. To the best of our knowledge, this is the first time that a quantified analysis of the capabilities of optical devices to mimic state-of-the-art activation functions is presented. The obtained activation functions are benchmarked on two machine learning examples: classification task using the Iris dataset, and image recognition using the MNIST dataset. We use complex-valued feed-forward neural networks and get test accuracies of 97% and 95% respectively.

show abstract

Section: Appendix B Scalabilitymentioning

confidence: 99%

Reconfigurable Activation Functions in Integrated Optical Neural Networks

Campo¹,

Pérez²

2022

IEEE J. Select. Topics Quantum Electron.

View full text Add to dashboard Cite

show abstract

“…The S2D stage converts the single-ended TIA output into a differential signal using an inverter loaded by a matched diode-connected inverter [11], as shown in Fig. 5.…”

Section: A Multi-stage Sf-tia and S2dmentioning

confidence: 99%

“…Because TIA output has a common-mode voltage of V DD /2, implementing the VGA-GM using a CMOS input differential pair achieves better linearity performance with large input signal swing compared to PMOS-or NMOS-only differential-pair-based VGAs as in [5] and [16]. However, tail current sources reduce the voltage headroom, compared to pseudo-differential GM stage in [11], but they help improve the common-mode rejection and increase immunity to supply variations specially at low frequencies. Source degeneration resistors (R D1 and R D2 ) are used to enhance the VGA linearity performance.…”

Section: B Vgamentioning

confidence: 99%

“…Given these drawbacks, this article presents a high-sensitivity low-power optical RX using duobinary sampling in conjunction with a low-bandwidth TIA. Instead of canceling the ISI arising from the low-bandwidth TIA, duobinary sampling that leverages the well-controlled ISI [10], resulting from well-defined frequency response [11], is used to improve RX sensitivity. Low-latency feedback loops are automatically eliminated due to the duobinary data recovery; error propagation is mitigated by using simple pre-coding.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A 16-Gb/s -11.6-dBm OMA Sensitivity 0.7-pJ/bit Optical Receiver in 65-nm CMOS Enabled by Duobinary Sampling

Ahmed

Kim

Nandwana

et al. 2021

IEEE J. Solid-State Circuits

Self Cite

View full text Add to dashboard Cite

High-speed, low-power optical interconnects, such as intensity modulation direct detection (IMDD) optical links, are increasingly deployed in data centers to keep pace with the growing bandwidth requirements. High-sensitivity low-power optical receivers (RXs) are the key components that enable energy-efficient IMDD optical interconnects. This article presents a low-power nonreturn-to-zero (NRZ) optical RX using a combination of a limited-bandwidth trans-impedance amplifier (TIA) and duobinary sampling to improve RX sensitivity at high data rates. Duobinary sampling leverages the well-controlled TIA inter-symbol interference (ISI) to recover the transmitted data, making it much more hardware efficient than canceling the ISI using a decision feedback equalizer (DFE). The proposed optical RX employs a CMOS-based analog front-end (AFE) to achieve high linearity and excellent power efficiency. Fabricated in 65-nm CMOS process, the prototype RX achieves optical modulation amplitude (OMA) sensitivity of −11.6 dBm at 16 Gb/s with 0.7-pJ/bit efficiency.

show abstract

“…Thus, we assume similar energy efficiency for multilevel signaling as PAM2, ∼ 2pJ/b, for MZM modulators 25 . For binary resolution, the power consumed by the drivers and AFEs are extracted from recent work on PAM2 transceivers 5,25,62,65 . Therefore, the energy efficiency for 2b, 3b and 4b input MZM drivers are estimated in our calculation as ∼ 4pJ, 6pJ and 8pJ per symbol, respectively.…”

Section: Efficiency Tradeoff Factorsmentioning

confidence: 99%

Scaling Up Silicon Photonic-based Accelerators: Challenges and Opportunities

Al-Qadasi,

Chrostowski,

Shastri

et al. 2021

Preprint

View full text Add to dashboard Cite

A Process and Temperature Insensitive CMOS Linear TIA for 100 Gb/s/$\lambda$ PAM-4 Optical Links

Cited by 47 publications

References 7 publications

Reconfigurable Activation Functions in Integrated Optical Neural Networks

Reconfigurable Activation Functions in Integrated Optical Neural Networks

A 16-Gb/s -11.6-dBm OMA Sensitivity 0.7-pJ/bit Optical Receiver in 65-nm CMOS Enabled by Duobinary Sampling

Scaling Up Silicon Photonic-based Accelerators: Challenges and Opportunities

Contact Info

Product

Resources

About