As process technology is scaled down, a typical system on a chip (SoC) becomes denser. In scaled process technology, process variation becomes greater and increasingly affects the SoC circuits. Moreover, the process variation strongly affects network-on-chips (NoCs) that have a synchronous network across the chip. Therefore, its network frequency is degraded. We propose a process-variation-adaptive NoC with a variation-adaptive variable-cycle router (VAVCR). The proposed VAVCR can configure its cycle latency adaptively on a processor core basis, corresponding to the process variation. It can increase the network frequency, which is limited by the process variation in a conventional router. Furthermore, we propose a variable-cycle pipeline adaptive routing (VCPAR) method with VAVCR; the proposed VCPAR can reduce packet latency and has tolerance to network congestion. The total execution time reduction of the proposed VAVCR with VCPAR is 15.7%, on average, for five task graphs.

This paper proposes 7T SRAM which realizes block-level simultaneous copying feature. The proposed SRAM can be used for data transfer between local memories such as checkpoint data storage and transactional memory. The 1-Mb SRAM is comprised of 32-kb blocks, in which 16-kb data can be copied in 33.3 ns at 1.2 V. The proposed scheme reduces energy consumption in copying by 92.7% compared to the conventional read-modify-write manner. By applying the proposed scheme to transactional memory, the number of write back cycles is possibly reduced by 98.7% compared with the conventional memory system.

This paper presents a resilient cache memory for dynamic variation tolerance in a 40-nm CMOS. The cache can perform sustained operations under a large-amplitude voltage droop. To realize sustained operation, the resilient cache exploits 7T/14T bit-enhancing SRAM and on-chip voltage/temperature monitoring circuit. 7T/14T bit-enhancing SRAM can reconfigure itself dynamically to a reliable bit-enhancing mode. The on-chip voltage/temperature monitoring circuit can sense a precise supply voltage level of a power rail of the cache. The proposed cache can dynamically change its operation mode using the voltage/temperature monitoring result and can operate reliably under a large-amplitude voltage droop. Experimental result shows that it does not fail with 25% and 30% droop of Vdd and it provides 91 times better failure rate with a 35% droop of Vdd compared with the conventional design.

This paper presents a novel architecture for a fault-tolerant and dual modular redundancy (DMR) system using a checkpoint recovery approach. The architecture features exploitation of SRAM with simultaneous copy and instantaneous compare function. It can perform low-latency data copying between dual cores. Therefore, it can carry out fast backup and rollback. Furthermore, it can reduce the power consumption during data comparison process compared to the cyclic redundancy check (CRC). Evaluation results show that, compared with the conventional checkpoint/restart DMR, the proposed architecture reduces the cycle overhead by 97.8% and achieves a 3.28% low-latency execution cycle even if a one-time fault occurs when executing the task. The proposed architecture provides high reliability for systems with a real-time requirement.

This paper presents an adaptive cache architecture for wide-range reliable low-voltage operations. The proposed associativity-reconfigurable cache consists of pairs of cache ways so that it can exploit the recovery feature of the novel 7T/14T SRAM cell. Each pair has two operating modes that can be selected based upon the required voltage level of current operating conditions: normal mode for high performance and dependable mode for reliable low-voltage operations. We can obtain reliable low-voltage operations by application of the dependable mode to weaker pairs that cannot operate reliably at low voltages. Meanwhile leaving stronger pairs in the normal mode, we can minimize performance losses. Our chip measurement results show that the proposed cache can trade off its associativity with the minimum operating voltage. Moreover, it can decrease the minimum operating voltage by 140 mV achieving 67.48% and 26.70% reduction of the power dissipation and energy per instruction. Processor simulation results show that designing the on-chip caches using the proposed scheme results in 2.95% maximum IPC losses, but it can be chosen various performance levels. Area estimation results show that the proposed cache adds area overhead of 1.61% and 5.49% in 32-KB and 256-KB caches, respectively.