Techn. Univ., Diss.--Ilmenau, 2006.

The electromigration (EM) and chip-package interaction (CPI) reliability of flip chip packages with Cu pillar structures was investigated. First the EM-related characteristics of Cu pillars with solder tips were studied and compared with standard controlled collapse chip connection (C4) Pb-free solder joints. The simulation results revealed a significant reduction in the current crowding effect when C4 solder joints was replaced by Cu pillar structures. As a result, the current-induced Joule heating and local temperature gradients were reduced in the Cu pillar structure. This was followed by a study of the impact of the Cu pillar bumps on the mechanical reliability of low-k dielectrics. The CPI-induced crack driving force for delamination in the low-k interconnect structure was evaluated using a 3D sub-modeling technique. The energy release rate was found to increase significantly for packages with Cu pillar bumps compared with those with C4 Pb-free solder joints only. Structural optimization of Cu pillar bumps to improve the mechanical stability of packages with low-k chips was discussed. / text

Decisions regarding the mapping of system-on-chip (SoC) components onto a NoC become more difficult with increasing complexity of system design. These complex systems capable of providing multiple functionalities tend to operate in multiple modes of operation. Modeling the system communication in these multimodes aids in efficient system design. This research provides a heuristic that gives a flexible mapping solution of the multimode system communications onto the NoC topology of choice. The solution specifies the immediate neighbors of the SoC components and the routes taken by all communications in the system. We validate the mapping results with a network-on-chip simulator (NoCSim). This thesis also investigates the cost associated with the interfacing of the components to the NoC. With the goal of reducing communication latency, we examine the packetization strategies in the NoC communication. Three schemes of implementations were analyzed, and the costs in terms of latency, and area were projected through actual synthesis.

In this dissertation, I explore energy and reliability in future NoC (Network-on-Chip) interconnected CMPs (chip multiprocessors) as they have become a first-order constraint in future CMP design. In the first part, we target the root cause of network energy consumption through techniques that reduce link and router-level switching activity. We specifically focus on memory subsystem traffic, as it comprises the bulk of NoC load in a CMP. By transmitting only the flits that contain words that we predicted would be useful using a novel spatial locality predictor, our scheme seeks to reduce network activity. We aim to further lower NoC energy consumption through microarchitectural mechanisms that inhibit datapath switching activity caused by unused words in individual flits. Using simulation-based performance studies and detailed energy models based on synthesized router designs and different link wire types, we show that (a) the pre- diction mechanism achieves very high accuracy, with an average rate of false-unused prediction of just 2.5%; (b) the combined NoC energy savings enabled by the predictor and microarchitectural support are 36% on average and up to 57% in the best case; and (c) there is no system performance penalty as a result of this technique. In the second part, we present a method for dynamic voltage/frequency scaling of networks-on-chip and last level caches in CMP designs, where the shared resources form a single voltage/frequency domain. We develop a new technique for monitoring and control and validate it by running PARSEC benchmarks through full system simulations. These techniques reduce energy-delay product by 46% compared to a state-of-the-art prior work. In the third part, we develop critical path models for HCI- and NBTI-induced wear assuming stress caused under realistic workload conditions, and apply them onto the interconnect microarchitecture. A key finding from this modeling is that, counter to prevailing wisdom, wearout in the CMP on-chip interconnect is correlated with a lack of load observed in the NoC routers, rather than high load. We then develop a novel wearout-decelerating scheme in which routers under low load have their wearout-sensitive components exercised without significantly impacting the router’s cycle time, pipeline depth, and area or power consumption. We subsequently show that the proposed design yields a 13.8∼65× increase in CMP lifetime.

With increased density of modern System on Chip(SoC) communication between nodes has become a major problem. Network on Chip is a novel on chip communication paradigm to solve this by using highly scalable and efficient packet switched network. The addition of intelligent networking on the chip adds to the chip’s power consumption thus making management of communication power an interesting and challenging research problem. While VLSI techniques have evolved over time to enable power reduction in the circuit level, the highly dynamic nature of modern large SoC demand more than that. This dissertation explores some innovative dynamic solutions to manage the ever increasing communication power in the post sub-micron era. Today’s highly integrated SoCs require great level of cross layer optimizations to provide maximum efficiency. This dissertation aims at the dynamic power management problem from top. Starting with a system level distribution and management down to microarchitecture enhancements were found necessary to deliver maximum power efficiency. A distributed power budget sharing technique is proposed. To efficiently satisfy the established power budget, a novel flow control and throttling technique is proposed. Finally power efficiency of underlying microarchitecture is explored and novel buffer and link management techniques are developed. All of the proposed techniques yield improvement in power-performance efficiency of the NoC infrastructure.

As modern CMPs scale to ever increasing core counts, Networks-on-Chip (NoCs) are emerging as an interconnection fabric, enabling communication between components. While NoCs are easy to implement and provide high and scalable bandwidth, current routing algorithms, such as dimension-ordered routing, suffer from poor load balance, leading to reduced throughput and high latencies. Improving load balance, hence, is critical in future CMP designs where increased latency leads to wasted power and energy waiting for outstanding requests to resolve. Adaptive routing is a known technique to improve load balance; however, prior adaptive routing techniques either use local, myopic information or misinformed, regionally-aggregated information to form their routing decisions. This thesis proposes a new, light-weight, adaptive routing algorithm for on-chip routers based on global link state and congestion information, Global Congestion Awareness (GCA). GCA leverages unused bits in existing packet header flits to "piggyback" congestion state information around the network and uses a simple, low-complexity route calculation unit, to calculate optimal packet paths to their destination without the myopia of local decisions, nor the aggregation of unrelated status information, found in prior designs. In particular GCA outperforms local adaptive routing by up to 82%, Regional Congestion Awareness (RCA) by up to 51%, and a recent competing adaptive routing algorithm, DAR, by 8% on average on realistic workloads.

O consumo de potência em uma Rede Intra-Chip (em inglês, Network-on-Chip – NoC) cresce linearmente com a quantidade de transições de sinais nos pacotes transmitidos através da infra-estrutura de interconexão. Uma forma de minimizar o consumo de potência em um sistema baseado em NoC é reduzir a atividade de transição de sinais nas portas de entrada dos módulos que constituem a NoC. Esta redução pode ser obtida através da utilização de esquemas de codificação de dados. Vários esquemas de codificação foram propostos no final dos anos 90, porém direcionados a arquiteturas de comunicação baseadas em barramentos. Este trabalho investiga a utilização destes esquemas de codificação em sistemas baseados em Networks-on-Chip. Dentre os esquemas encontrados na literatura, quatro foram implementados e avaliados neste trabalho. Este trabalho também apresenta como contribuição original um novo esquema de codificação de dados adequado a NoCs. A estimativa do consumo de potência da NoC é calculada com base em macromodelos que reproduzem a potência consumida em cada módulo interno da NoC, de acordo com a atividade de transição de sinais no tráfego recebido. Estes macromodelos são aqui caracterizados através da simulação elétrica de cada módulo da NoC e dos esquemas de codificação. Para permitir a análise de consumo com tráfegos de aplicações reais, os macromodelos são inseridos em um modelo de mais alto nível de abstração. Este modelo é empregado para analisar o balanço entre redução de potência obtida com a redução da transição de sinais e o consumo extra do esquema de codificação. A maioria dos esquemas de codificação encontrados na literatura reduz efetivamente a atividade de transição de sinais. Porém, o impacto do consumo extra de potência para codificar e decodificar os dados não é avaliado. A avaliação conduzida neste trabalho considera o consumo da codificação/decodificação em uma NoC real, quantificando a redução de consumo obtido com cada esquema de codificação. Devido ao baixo desempenho dos esquemas de codificação existentes, quando aplicados a NoCs, foi desenvolvido um novo esquema, chamado T-Bus-Invert. Os resultados mostram um desempenho superior do T-Bus-Invert quando comparado aos demais esquemas para flits com largura de 8 e 16 bits, e um desempenho similar ao do Bus-Invert com 4 clusters para flits de 32 bits. / The power consumption in Networks-on-Chip grows linearly with the amount of signal transitions in successive data packets sent through this interconnection infrastructure. One option to decrease the power consumption in NoC-based systems is reducing the switching activity at the input ports of NoC modules. This reduction can be achieved by means of data coding schemes. Several schemes were proposed in the nineties. However, all of them address only bus-based communication architectures. This work investigates the use of such data coding schemes in NoC-based systems. Among the coding schemes found in the literature, four were implemented and evaluated in this work. This work also presents a new data coding scheme, named TBus- Invert, suitable for NoCs. Estimations of the NoC power consumption are computed here based on macromodels which reproduce the power consumption on each internal NoC module, according to the transition activity in the input traffic. Such macromodels are characterized through electrical simulations of each NoC module and coding circuits. To enable the evaluation of real applications traffic, such macromodels are inserted in a higher abstraction level model. This model is employed to analyze the trade-off between the power saving due to coding schemes versus the power consumption overhead due to the encoding and decoding modules. Most of the coding schemes proposed in the literature effectively reduce the switching activity, but the overall impact of the power consumption to encode/decode data in the system is not evaluated. The evaluation conducted in this work considers the power consumption to encode/decode data in a real NoC, quantifying the power savings for each coding scheme. Due to the insufficient performances of the existing schemes when applied to NoCs, a coding scheme, T-Bus-Invert, was developed. Results showed superior performance of the T-Bus-Invert compared to all evaluated coding schemes for 8 and 16-bit flits, and similar performance to the 4-cluster Bus-Invert for 32-bit flits.

O consumo de potência em uma Rede Intra-Chip (em inglês, Network-on-Chip – NoC) cresce linearmente com a quantidade de transições de sinais nos pacotes transmitidos através da infra-estrutura de interconexão. Uma forma de minimizar o consumo de potência em um sistema baseado em NoC é reduzir a atividade de transição de sinais nas portas de entrada dos módulos que constituem a NoC. Esta redução pode ser obtida através da utilização de esquemas de codificação de dados. Vários esquemas de codificação foram propostos no final dos anos 90, porém direcionados a arquiteturas de comunicação baseadas em barramentos. Este trabalho investiga a utilização destes esquemas de codificação em sistemas baseados em Networks-on-Chip. Dentre os esquemas encontrados na literatura, quatro foram implementados e avaliados neste trabalho. Este trabalho também apresenta como contribuição original um novo esquema de codificação de dados adequado a NoCs. A estimativa do consumo de potência da NoC é calculada com base em macromodelos que reproduzem a potência consumida em cada módulo interno da NoC, de acordo com a atividade de transição de sinais no tráfego recebido. Estes macromodelos são aqui caracterizados através da simulação elétrica de cada módulo da NoC e dos esquemas de codificação. Para permitir a análise de consumo com tráfegos de aplicações reais, os macromodelos são inseridos em um modelo de mais alto nível de abstração. Este modelo é empregado para analisar o balanço entre redução de potência obtida com a redução da transição de sinais e o consumo extra do esquema de codificação. A maioria dos esquemas de codificação encontrados na literatura reduz efetivamente a atividade de transição de sinais. Porém, o impacto do consumo extra de potência para codificar e decodificar os dados não é avaliado. A avaliação conduzida neste trabalho considera o consumo da codificação/decodificação em uma NoC real, quantificando a redução de consumo obtido com cada esquema de codificação. Devido ao baixo desempenho dos esquemas de codificação existentes, quando aplicados a NoCs, foi desenvolvido um novo esquema, chamado T-Bus-Invert. Os resultados mostram um desempenho superior do T-Bus-Invert quando comparado aos demais esquemas para flits com largura de 8 e 16 bits, e um desempenho similar ao do Bus-Invert com 4 clusters para flits de 32 bits. / The power consumption in Networks-on-Chip grows linearly with the amount of signal transitions in successive data packets sent through this interconnection infrastructure. One option to decrease the power consumption in NoC-based systems is reducing the switching activity at the input ports of NoC modules. This reduction can be achieved by means of data coding schemes. Several schemes were proposed in the nineties. However, all of them address only bus-based communication architectures. This work investigates the use of such data coding schemes in NoC-based systems. Among the coding schemes found in the literature, four were implemented and evaluated in this work. This work also presents a new data coding scheme, named TBus- Invert, suitable for NoCs. Estimations of the NoC power consumption are computed here based on macromodels which reproduce the power consumption on each internal NoC module, according to the transition activity in the input traffic. Such macromodels are characterized through electrical simulations of each NoC module and coding circuits. To enable the evaluation of real applications traffic, such macromodels are inserted in a higher abstraction level model. This model is employed to analyze the trade-off between the power saving due to coding schemes versus the power consumption overhead due to the encoding and decoding modules. Most of the coding schemes proposed in the literature effectively reduce the switching activity, but the overall impact of the power consumption to encode/decode data in the system is not evaluated. The evaluation conducted in this work considers the power consumption to encode/decode data in a real NoC, quantifying the power savings for each coding scheme. Due to the insufficient performances of the existing schemes when applied to NoCs, a coding scheme, T-Bus-Invert, was developed. Results showed superior performance of the T-Bus-Invert compared to all evaluated coding schemes for 8 and 16-bit flits, and similar performance to the 4-cluster Bus-Invert for 32-bit flits.

O consumo de potência em uma Rede Intra-Chip (em inglês, Network-on-Chip – NoC) cresce linearmente com a quantidade de transições de sinais nos pacotes transmitidos através da infra-estrutura de interconexão. Uma forma de minimizar o consumo de potência em um sistema baseado em NoC é reduzir a atividade de transição de sinais nas portas de entrada dos módulos que constituem a NoC. Esta redução pode ser obtida através da utilização de esquemas de codificação de dados. Vários esquemas de codificação foram propostos no final dos anos 90, porém direcionados a arquiteturas de comunicação baseadas em barramentos. Este trabalho investiga a utilização destes esquemas de codificação em sistemas baseados em Networks-on-Chip. Dentre os esquemas encontrados na literatura, quatro foram implementados e avaliados neste trabalho. Este trabalho também apresenta como contribuição original um novo esquema de codificação de dados adequado a NoCs. A estimativa do consumo de potência da NoC é calculada com base em macromodelos que reproduzem a potência consumida em cada módulo interno da NoC, de acordo com a atividade de transição de sinais no tráfego recebido. Estes macromodelos são aqui caracterizados através da simulação elétrica de cada módulo da NoC e dos esquemas de codificação. Para permitir a análise de consumo com tráfegos de aplicações reais, os macromodelos são inseridos em um modelo de mais alto nível de abstração. Este modelo é empregado para analisar o balanço entre redução de potência obtida com a redução da transição de sinais e o consumo extra do esquema de codificação. A maioria dos esquemas de codificação encontrados na literatura reduz efetivamente a atividade de transição de sinais. Porém, o impacto do consumo extra de potência para codificar e decodificar os dados não é avaliado. A avaliação conduzida neste trabalho considera o consumo da codificação/decodificação em uma NoC real, quantificando a redução de consumo obtido com cada esquema de codificação. Devido ao baixo desempenho dos esquemas de codificação existentes, quando aplicados a NoCs, foi desenvolvido um novo esquema, chamado T-Bus-Invert. Os resultados mostram um desempenho superior do T-Bus-Invert quando comparado aos demais esquemas para flits com largura de 8 e 16 bits, e um desempenho similar ao do Bus-Invert com 4 clusters para flits de 32 bits. / The power consumption in Networks-on-Chip grows linearly with the amount of signal transitions in successive data packets sent through this interconnection infrastructure. One option to decrease the power consumption in NoC-based systems is reducing the switching activity at the input ports of NoC modules. This reduction can be achieved by means of data coding schemes. Several schemes were proposed in the nineties. However, all of them address only bus-based communication architectures. This work investigates the use of such data coding schemes in NoC-based systems. Among the coding schemes found in the literature, four were implemented and evaluated in this work. This work also presents a new data coding scheme, named TBus- Invert, suitable for NoCs. Estimations of the NoC power consumption are computed here based on macromodels which reproduce the power consumption on each internal NoC module, according to the transition activity in the input traffic. Such macromodels are characterized through electrical simulations of each NoC module and coding circuits. To enable the evaluation of real applications traffic, such macromodels are inserted in a higher abstraction level model. This model is employed to analyze the trade-off between the power saving due to coding schemes versus the power consumption overhead due to the encoding and decoding modules. Most of the coding schemes proposed in the literature effectively reduce the switching activity, but the overall impact of the power consumption to encode/decode data in the system is not evaluated. The evaluation conducted in this work considers the power consumption to encode/decode data in a real NoC, quantifying the power savings for each coding scheme. Due to the insufficient performances of the existing schemes when applied to NoCs, a coding scheme, T-Bus-Invert, was developed. Results showed superior performance of the T-Bus-Invert compared to all evaluated coding schemes for 8 and 16-bit flits, and similar performance to the 4-cluster Bus-Invert for 32-bit flits.

Com o crescimento acelerado da complexidade das aplicações e softwares que exigem alto desempenho, o hardware e sua arquitetura passou por algumas mudanças para que pudesse atender essa necessidade. Uma das abordagens propostas e desenvolvidas para suportar essas aplicações, foi a integração de mais de um core de processamento em um único circuito integrado. Inicialmente, a comunicação utilizando barramento foi escolhida, pela sua vantagem de reuso comparado a ponto a ponto. No entanto, com o aumento acelerado da quantidade de cores nos Systems-on-Chip (SoC), essa abordagem passou a apresentar problemas para suportar a comunicação interna. Uma alternativa que vem sendo explorada é a Network-on-Chip (NoC), uma abordagem que propõe utilizar o conhecimento de redes comuns em projetos de comunicação interna de SoC. Esse trabalho fornece uma arquitetura de NoC completa, configurável, parametrizável e no padrão Ethernet. Os três módulos básicos da NoC, Network Adapter (NA), Link e Switch, são implementados e disponibilizados. Os resultados foram obtidos utilizando o FPGA Stratix IV da Altera. As métricas de desempenho utilizadas para validação da NoC são a área no FPGA e o atraso na comunicação. Os parâmetros disponibilizados são referentes as configurações dos módulos desenvolvidos, considerando características apresentadas de aplicações DSP (Digital Signal Processing). O experimento utilizando dois NAs, dois cores e um Switch precisou de 7310 ALUTs do FPGA EP4SGX230KF40C2ES o que corresponde a 4% dos seus recursos lógicos. O tempo gasto para a transmissão de um quadro ethernet de 64 Bytes foi de 422 ciclos de clock a uma frequência de 50MHz. / With the accelerated growth of the complexity of the software and applications that require high performance, hardware and its architecture has undergone a few changes so it could meet that need. One of the proposals and approaches developed to support these applications, was the integration of more than one core processing in a single integrated circuit. Initially, the bus communication architecture was chosen, using for its reuse benefit compared to point-to-point. However, with the cores number increase in Systems-on-Chip (SoC), this approach began to present problems to support internal communication. An alternative that has been explored is the Network-on-Chip (NoC), an approach that proposes to use knowledge of common networks on internal communication projects of SOC. This dissertation focuses is to provide a complete NoC architecture, configurable, customizable and on standard Ethernet. The three NoC basic modules, Network Adapter (NA), Link and Switch, are implemented. The results were obtained using the Stratix IV FPGA. The performance metrics used for NoC validation are silicon area and latency. The available parameters are related to developed modules settings, considering features presented of DSP applications. The experiment using two NA, two cores and one Switch needed 7310 FPGA ALUTs which corresponds to 4% of their logical resources. The time for the transmission of an ethernet frame of 64 Bytes was 422 clock cycles at 50 MHz.