Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
Publication
Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
Over the past few decades, incessant growth of Internet networking traffic and High-Performance Computing (HPC) has led to a tremendous demand for data bandwidth. Digital communication technologies combined with advanced integrated circuit scaling trends have enabled the semiconductor and microelectronic industry to dramatically scale the bandwidth of high-loss interfaces such as Ethernet, backplane, and Digital Subscriber Line (DSL). The key to achieving higher bandwidth is to employ equalization technique to compensate the channel impairments such as Inter-Symbol Interference (ISI), crosstalk, and environmental noise. Therefore, todayâs advanced input/outputs (I/Os) has been equipped with sophisticated equalization techniques to push beyond the uncompensated bandwidth of the system. To this end, process scaling has continually increased the data processing capability and improved the I/O performance over the last 15 years. However, since the channel bandwidth has not scaled with the same pace, the required signal processing and equalization circuitry becomes more and more complicated. Thereby, the energy efficiency improvements are largely offset by the energy needed to compensate channel impairments. In this design paradigm, re-thinking about the design strategies in order to not only satisfy the bandwidth performance, but also to improve power-performance becomes an important necessity. It is well known in communication theory that coding and signaling schemes have the potential to provide superior performance over band-limited channels. However, the choice of the optimum data communication algorithm should be considered by accounting for the circuit level power-performance trade-offs. In this thesis we have investigated the application of new algorithm and signaling schemes in wireline communications, especially for communication between microprocessors, memories, and peripherals. A new hybrid NRZ/Multi-Tone (NRZ/MT) signaling method has been developed during the course of this research. The system-level and circuit-level analysis, design, and implementation of the proposed signaling method has been performed in the frame of this work, and the silicon measurement results have proved the efficiency and the robustness of the proposed signaling methodology for wireline interfaces. In the first part of this work, a 7.5 Gb/s hybrid NRZ/MT transceiver (TRX) for multi-drop bus (MDB) memory interfaces is designed and fabricated in 40 nm CMOS technology. Reducing the complexity of the equalization circuitry on the receiver (RX) side, the proposed architecture achieves 1 pJ/bit link efficiency for a MDB channel bearing 45 dB loss at 2.5 GHz. The measurement results of the first prototype confirm that NRZ/MT serial data TRX can offer an energy-efficient solution for MDB memory interfaces. Motivated by the satisfying results of the first prototype, in the second phase of this research we have exploited the properties of multi-tone signaling, especially orthogonality among different sub-bands, to reduce the effect of crosstalk in high-dense wireline interconnects. A four-channel transceiver has been implemented in a standard CMOS 40 nm technology in order to demonstrate the performance of NRZ/MT signaling in presence of high channel loss and strong crosstalk noise. The proposed system achieves 1 pJ/bit power efficiency, while communicating over a MDB memory channel at 36 Gb/s aggregate data rate.