Low-Power Embedded Design Solutions and Low-Latency On-Chip Interconnect Architecture for System-On-Chip Design

This dissertation presents three design solutions to support several key system-on-chip (SoC) issues to achieve low-power and high performance. These are: 1) joint source and channel decoding (JSCD) schemes for low-power SoCs used in portable multimedia systems, 2) efficient on-chip interconnect architecture for massive multimedia data streaming on multiprocessor SoCs (MPSoCs), and 3) data processing architecture for low-power SoCs in distributed sensor network (DSS) systems and its implementation. The first part includes a low-power embedded low density parity check code (LDPC) - H.264 joint decoding architecture to lower the baseband energy consumption of a channel decoder using joint source decoding and dynamic voltage and frequency scaling (DVFS). A low-power multiple-input multiple-output (MIMO) and H.264 video joint detector/decoder design that minimizes energy for portable, wireless embedded systems is also designed. In the second part, a link-level quality of service (QoS) scheme using unequal error protection (UEP) for low-power network-on-chip (NoC) and low latency on-chip network designs for MPSoCs is proposed. This part contains WaveSync, a low-latency focused network-on-chip architecture for globally-asynchronous locally-synchronous (GALS) designs and a simultaneous dual-path routing (SDPR) scheme utilizing path diversity present in typical mesh topology network-on-chips. SDPR is akin to having a higher link width but without the significant hardware overhead associated with simple bus width scaling. The last part shows data processing unit designs for embedded SoCs. We propose a data processing and control logic design for a new radiation detection sensor system generating data at or above Peta-bits-per-second level. Implementation results show that the intended clock rate is achieved within the power target of less than 200mW. We also present a digital signal processing (DSP) accelerator supporting configurable MAC, FFT, FIR, and 3-D cross product operations for embedded SoCs. It consumes 12.35mW along with 0.167mm2 area at 333MHz