Using PIM to overcome memory bottleneck • Although various bandwidth increase methods have been proposed, it is physically impossible to achieve a breakthrough increase. - Limited by # of PCB wires, # of CPU ball, and thermal constraints • PIM has been proposed to improve performance of bandwidth-intensive workloads and improve energy efficiency by reducing computing-memory data movement.

Technology scaling has continuously improved the density, performance, energy efficiency, and cost of DRAM-based main memory systems. Starting from sub-20nm processes, however, the industry began to pay considerably higher costs to screen and manage notably increasing defective cells. The traditional technique, which replaces the rows/columns containing faulty cells with spare rows/columns, has been able to cost-effectively repair the defective cells so far, but it will become unaffordable soon because an excessive number of spare rows/columns are required to manage the increasing number of defective cells. This necessitates a synergistic application of an alternative resilience technique such as In-DRAM ECC with the traditional one. Through extensive measurement and simulation, we first identify that aggressive miniaturization makes DRAM cells more sensitive to random telegraph noise or variable retention time, which is dominantly manifested as a surge in randomly scattered single-cell faults. Second, we advocate using InDRAM ECC to overcome the DRAM scaling challenges and architect In-DRAM ECC to accomplish high area efficiency and minimal performance degradation. Moreover, we show that advancement in process technology reduces decoding/correction time to a small fraction of DRAM access time, and that the throughput penalty of a write operation due to an additional read for a parity update is mostly overcome by the multi-bank structure and long burst writes that span an entire In-DRAM ECC codeword. Lastly, we demonstrate that system reliability with modern rank-level ECC schemes such as single device data correction is further improved by hundred million times with the proposed In-DRAM ECC architecture.

3D DRAM is the next-generation memory system targeting high bandwidth, low power, and small form factor. This paper presents a cross-domain CAD/architectural platform that addresses DC power noise issues in 3D DRAM targeting stacked DDR3, Wide I/O, and hybrid memory cube technologies. Our design and analysis include both individual DRAM dies and a host logic die that communicates with them in the same stack. Moreover, our comprehensive solutions encompass all major factors in design, packaging, and architecture domains, including power delivery network wire sizing, redistribution layer routing, distributed, and dedicated TSV placement, die bonding style, backside wire bonding, and read policy optimization. We conduct regression analysis and optimization to obtain high quality solutions under noise, cost, and performance tradeoff. Compared with industry standard baseline designs and policies, our methods achieve up to 68.2% IR-drop reduction and 30.6% performance enhancement.

This paper proposes PVT-invariant single-to-differential signal converter (SDQ applicable to the output circuitry of high-speed DDR SDRAM. The proposed SDC generates PVT-invariant differential-output sampling clock using a phase interpolation technique and a symmetric structure, and improves the aperture window of output data in source synchronous DDR SDRAM. The proposed SDC was simulated using 1.8-V 80-nm DRAM technology. The comparison result indicates that the differential clocks generated by the proposed SDC achieve 80.6% reduction of skew, 76.6% reduction of duty-cycle distortion, 61.7% of reduction of delay variation, and 8.5% reduction of maximum current for a given process, voltage, and temperature (PVT) variations, as compared to conventional SDCs. The I/O interface of a source-synchronous DDR SDRAM designed using the proposed SDC, which is operating at 1.0-Gbps/pin data rate has aperture window increased by 15.3% and ISI improved by 67.7% in comparison to conventional I/O interface.

This paper presents a mesh with virtual buses as the bandwidth-efficient implementation of the mesh with multiple broadcasting on which many computational problems can be solved with reduced time complexity. The new system provides a low latency and high bandwidth communication mechanism without the disadvantage of the dual network approach: the possible bandwidth waste. The virtual buses are built using those communication links of the base point-to-point mesh network only upon request. Even if bus requests are not made, the full network bandwidth is harnessed by the point-to-point communication. On the contrary, the bandwidth assigned to the conventional real buses will be wasted when there is little bus request. This paper introduces a wormhole router design equipped with the virtual bus functions, describes the connections among these routers, and presents various virtual bus transactions. We prove the effectiveness of the virtual bus by showing that a representative semigroup computation can be solved very efficiently on it: a vector maxima finding is accelerated by 2.66 times in the experiment. We also explore the network characteristics by the distribution-driven simulation of the system. These evaluations convinced us that the mesh with virtual buses is a promising approach to low latency and high bandwidth communication which has applications in many parallel computations such as in parallel simulations, computer graphics, real time systems, and many others.

This paper describes the MetaCore system which is an ASIP (Application-Specific Instruction set Processor) development system targeted for DSP applications. The goal of MetaCore system is to offer an efficient design methodology meeting specifications given as a combination of performance, cost and design turnaround time.

The performance and efficiency of event-driven simulations, such as VHDL and Verilog simulation, depend on the number of events that occur during the simulation. In this paper, we classify events into two categories, sensitive events and insensitive events, according to the necessity of simulations, and show classification algorithms for both combinational circuits and sequential circuits. We also implement the optimization methodology that eliminates unnecessary simulation activity caused by the insensitive events. VHDL programs can be rewritten by augmenting suppressed sensitivity lists. Experiments show that the optimized VHDL programs run almost two times faster than the original ones.