HPL(High Performance Linpack) is the standard benchmark used to evaluate supercomputers (high-performance computing systems) around the world. HPL solves a linear system of equations, Ax=b, through a series of mathematical processes such as 2D Block-Cyclic matrix distribution and LU Decomposition. In order to achieve the best performance, optimization of key HPL parameters is essential. In this paper, we propose an optimization analysis technique of HPL parameters based on HPLinpack performance results and the efficiency graph pattern on a homogeneous system consisting of the Neoverse N1SDP(System Development Platform) from ARM. Results show the significant influence of GEMM operations on performance and the need for its acceleration were further verified through HPL function call profiling. Finally, we show the effectiveness of the proposed parameter optimization methods on a heterogeneous system. A HPL modified for heterogenous system exhibited considerably improved performance when tested on the Neoverse N1SDP and the Nvidia RTX 2060 GPU.

In this demonstration, we present AB9 SoC system, a single-chip solution for AI application. It provides the reconfigurable and programmable architecture to support the general computations for a variety of neural networks. The AB9 SoC is implemented using TSMC 28-nm process technology with a chip size of 17×23 mm2and 1GHz operating frequency.

With fluctuating voltage, widening operating temperature, and increasing clock frequency, cache systems are becoming increasingly susceptible to transient (soft) errors. Error correction code(ECC) is an attractive approach for transient error detection and correction[1]. However, redundant memory for error correction code has a significant impact on cost and increases the transient error rates. This paper presents a fault tolerant cache system of vision processors for automotive. The cache system has the small redundant memory and decreases the transient error rates. We proposed the mechanism which increases the error recovery rate and is considered with data cache characteristics.

The advancement of process technology enables the integration of multiple cores featuring parallel processing. The requirement of extensive memory bandwidth puts a major performance bottleneck in multi-core architectures for media applications. While the parallel memory system is a viable solution to account for a large amount of memory transactions required by multiple cores, memory access conflicts caused by simultaneous accesses to an identical memory page by two or more cores limit the performance of multi-core architectures. We propose and evaluate the programmable memory address shuffler associated with the novel memory shuffling algorithm integrated in multi-core architectures with parallel memory system. The address shuffler efficiently translates the requested memory addresses into the shuffled addresses such that access conflicts diminish by analyzing the access pattern of the application. We demonstrate that the shuffling of sub-pages is represented by cyclic linked list which enables partial address shuffling with the minimal number of shuffling table entries. The programmable address shuffler reduces the amount of access conflicts by 83% for pitch-shifting audio decompression.

This paper describes the implementation of digital audio effect SoC (system on chip) which integrates the embedded DSP core, audio codec IP, a number of peripheral blocks and various audio effect algorithms. The audio effect SoC is developed by a software and hardware co-design method. The embedded DSP and some dedicated hardware blocks are developed as a hardware design while the audio effect algorithms are realized by a software centric way. In the implementation of audio effects, we employ the primitive functions of the embedded DSP compiler which provide the efficient way to implement the audio effect algorithms. Most of the algorithms are implemented by C language with the primitive functions and they are run on the embedded DSP while the equalizer which requires a large amount of computing power is implemented by the dedicated hardware block with high flexibility. The audio effect SoC was fabricated by using 0.18 um CMOS process and it was evaluated successfully on the real-time test board.

Programmability requirement in reconfigurable systems necessitates the integration of soft processors in FPGAs. The extensive memory bandwidth sets a major performance bottleneck in soft processors for media applications. While the parallel memory system is a viable solution to account for a large amount of memory transactions in media processors, the memory access conflicts caused by multiple memory buses limit the overall performance. We propose and evaluate the configurable memory address shuffler to be integrated in the memory access arbiter for the parallel memory system in a soft processor. The novel address shuffling algorithm reallocates the decomposed memory sub-pages based on the access conflict graph obtained by profiling the memory access pattern of the application to produce the synthesizable code. The address shuffler efficiently translates the requested memory addresses into the shuffled addresses such that the amount of simultaneous accesses to the identical physical memory block diminishes. The reconfigurability of the address shuffler enables the adaptive address shuffling depending on the memory access pattern of an application running on the soft processor. The configurable address shuffler reduces the amount of access conflicts by 80% on average utilizing 1592 LUTs which is 14% of that of the processor.

This paper presents an optimized channel usage between simulator and accelerator when the simulator models transaction-level SoC while accelerator models RTL sub-blocks. Conventional simulation accelerators synchronize the progresses of simulator and accelerator at every simulation time, which results in poor performance by splitting transactions on the simulator-to-accelerator channel into pieces. Occasional synchronization with predictions and recoveries makes it possible to merge multiple transfers yielding substantial performance gain compared to the conventional method.

Simulation is the most viable solution for the functional verification of system-on-chip (SoC). The acceleration of simulation with multi-field programmable gate array (multi-FPGA) emulator is a promising method to comply with the increasing complexity and large gate capacity of SoC. Time multiplexing of interconnection wires is the inevitable solution to solve the pin limitation problem that limits the gate utilization of FPGAs and speed of multi-FPGA simulation accelerators. The most time-consuming factor of multi-FPGA simulation acceleration is the synchronization time between a software simulator and a multi-FPGA system and the inter-FPGA synchronization time. This paper proposes a performance-driven signal synchronization mechanism for a simulation accelerator with multiple FPGAs using time-multiplexed interconnection. The event-based signal synchronization optimizes the synchronization time between a software simulator and the multi-FPGA system as well as the synchronization time among FPGAs. The synchronization time among FPGAs is optimized by circuit partitioning considering the signal probability, net dependency reduction, and efficient net clustering to reduce addressing overhead. The synchronization time between the software simulator and the multi-FPGA system is also optimized by exploiting the event probability of primary nets. Experiments show that the synchronization time is reduced to 6.2-9.8% of traditional mechanisms.

Logic emulation is so far the fastest method to verify the system functionality in the gate level before chip fabrication. Field-programmable gate array (FPGA)-based logic emulator with large gate capacity generally comprises a large number of FPGAs or special processors connected in mesh or crossbar topology. However, gate utilization of FPGAs and speed of emulation are limited by the number of signal pins among FPGAs and the interconnection architecture of the logic emulator. This paper first describes a new interconnection architecture called TOMi (Time-multiplexed, Off-chip, Multicasting interconnection) and proposes a circuit partitioning algorithm called ATOMi (Algorithm for TOMi) for multi-FPGA system incorporating four to eight FPGAs where FPGAs are interconnected through TOMi. ATOMi reduces the number of off-chip signal transfers to optimize the performance for multi-FPGA system implemented by TOMi. Experimental results using Partitioning93 benchmarks show that, by adopting the proposed TOMi interconnection architecture along with ATOMi, the pin count is reduced to 14.4%-88.6% while the critical path delay is reduced to 66.1%-90.1% compared to traditional architectures including mesh, crossbar, and VirtualWire architecture.

This paper presents new technology that accelerates system verification. Traditional methods for verifying functional designs are based on logic simulation, which becomes more time-consuming as design complexity increases. To accelerate functional simulation, hardware acceleration is used to offload calculation-intensive tasks from the software simulator. Hardware accelerated simulation dramatically reduces the simulation time. However, the communication overhead between the software simulator and hardware accelerator is becoming a new critical bottleneck. We reduce the communication overhead by exploiting burst data transfer and parallelism, which are obtained by splitting testbench and moving a part of testbench into hardware accelerator. Our experiments demonstrated that the proposed method reduces the communication overhead by a factor of about 40 compared to conventional hardware accelerated simulation while maintaining the cycle accuracy and compatibility with the original testbench.

Functional coverage is a technique for checking the completeness of test vectors in HDL simulation. Temporal events are used to monitor the sequence of events in the specification. In this paper, automatic generation of temporal events for functional coverage is proposed. The HiTER is the graph where nodes represent basic temporal properties or subgraph and edges represent time-shift value between two nodes. Hierarchical temporal events are generated by traversing HiTER such that invalid, or irrelevant properties are eliminated. Concurrent edge groups make it possible to generate more comprehensive temporal properties and hierarchical structure makes it easy to describe large design by combining multiple subgraphs. Automatically enerated temporal events describe almost all the possible temporal properties of the design under verification.

Functional coverage is a technique which can be used for checking the completeness of test vectors. In this paper, automatic generation of temporal events for functional coverage is proposed. The TERG (Temporal Event Relation Graph) is the graph where the nodes represent basic temporal property and the edges represent the time-shift value between two properties. Hierarchical temporal events are generated by traversing TERG such that invalid or irrelevant properties are eliminated. Concurrent edge groups in TERG make it possible to generate more comprehensive temporal properties.

FPGA-based logic emulator with large gate capacity generally comprises a large number of FPGAs connected in mesh or crossbar topology. However, gate utilization of FPGAs and speed of emulation are limited by the number of signal pins among FPGAs and the interconnection architecture of the logic emulator. The time-multiplexing of interconnection wires is required for multiFPGA system incorporating several state-of-the-art FPGAs. We propose a circuit partitioning algorithm called SCATOMi(scheduling driven algorithm for TOMi) for multiFPGA system incorporating four to eight FPGAs where FPGAs are interconnected through TOMi(time-multiplexed, off-chip, multicasting interconnection). SCATOMi improves the performance of TOMi architecture by limiting the number of inter-FPGA signal transfers on the critical path and considering the scheduling of inter-FPGA signal transfers. The performance of the partitioning result of SCATOMi is 5.5 times faster than traditional partitioning algorithms. Experiments on architecture comparison show that, by adopting the proposed TOMi interconnection architecture along with SCATOMi, the pin count is reduced to 15.2%-81.3% while the critical path delay is reduced to 46.1%-67.6% compared to traditional architectures including mesh, crossbar and VirtualWire architecture.

Abstract — This paper presents a special hardware implementation developed for the computation of the specular term which is the most time consuming part in the Phong's illumination. In the Phong shading, the exponentiation operation of two floating-point numbers is necessary for each point inside a polygon. An approximation algorithm is developed to speed up the exponentiation operation, and it is supported by simple hardware that can be easily merged into a floating-point multiplier. The exponentiation operation takes just 4 cycles in the proposed hardware while it takes about 100-200 cycles in conventional floating-point units. Although an approximation algorithm is employed for the exponentiation operation, the amount of error is so minimal that the difference is virtually indistinguishable. I.

Texture mapping is a common technique used to increase the visual quality of 3D scenes. As texture mapping requires a large amount of memory to deal with large textures generally required in the current visual systems, we propose an algorithm for compressing a pyramid texture used for mipmapping. Vector quantization is used to compress all levels of the pyramid texture to one representative value databook, one residual codebook and one index map. The proposed compression scheme uses interpolative texel difference vector quantization that compresses the difference between the interpolated surfaces generated by the representative value and the correct uncompressed texels of the texture at each level. The compressed pyramid texture can be accessed randomly and decompressed without loss of visual quality. We also propose a hardware architecture that performs the trilinear filtering with the compressed pyramid texture.

This paper describes the MetaCore system which is an application-specific instruction-set processor (ASIP) development system targeted for digital signal processor (DSP) applications. The goal of the MetaCore system is to offer an efficient design methodology meeting specifications given as a combination of performance, cost, and design turnaround time. The MetaCore system consists of two major design stages: design exploration and design generation. In the design exploration stage, MetaCore system accepts a set of benchmark programs and structural/behavioral specifications for the target processor and estimates the hardware cost and performance for each hardware configuration being explored. Once a hardware configuration and instruction set are chosen, the system helps generate the target processor design in the form of hardware description language (HDL) along with the application program development tools such as C compiler, assembler, and instruction set simulator. The effectiveness of the MetaCore system was verified with a successful design of MDSP-II, a programmable DSP processor targeted for mobile communication.

We propose a new flip-flop configuration which saves about 60% of total clocking power using a half-swing clock. To use the half-swing clock, level converters or special clock drivers are traditionally required and the power consumptions of this logic cannot be ignored. In the proposed scheme, only NMOS devices are clocked with a half-swing clock in order to make it operate without the level converter or any other additional logics, and the random logic circuits except for the clock and flip-flops are supplied by V/sub cc/ while the clock network is supplied by V/sub cc//2. Compared to the conventional scheme, a great amount of power consumed in clocking which is responsible for a large portion of total chip power can be saved with the proposed new flip-flop configuration.

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.