Parameter variations, noise susceptibility, and increasing energy dissipation of cmos devices have been recognized as major challenges in circuit and microarchitecture design in the nanometer regime. Among these, parameter variations and noise susceptibility are increasingly causing cmos devices to behave in an “unreliable” or “probabilistic” manner. To address these challenges, a shift in design paradigm from current-day deterministic designs to “statistical” or “probabilistic” designs is deemed inevitable. To respond to this need, in this article, we introduce and study an entirely novel family of probabilistic architectures: the probabilistic system-on-a-chip (psoc). psoc architectures are based on cmos devices rendered probabilistic due to noise, referred to as probabilistic CMOS or PCMOS devices. We demonstrate that in addition to harnessing the probabilistic behavior of pcmos devices, psoc architectures yield significant improvements, both in energy consumed as well as performance in the context of probabilistic or randomized applications with broad utility. All of our application and architectural savings are quantified using the product of the energy and performance, denoted (energy × performance): The pcmos-based gains are as high as a substantial multiplicative factor of over 560 when compared to a competing energy-efficient cmos-based realization. Our architectural design is application specific and involves navigating design space spanning the algorithm (application), its architecture (psoc), and the probabilistic technology (pcmos).

Probabilistic arithmetic, where the ith output bit of addition and multiplication is correct with a probability pi , is shown to be a vehicle for realizing extremely energy-efficient, embedded computing. Specifically, probabilistic adders and multipliers, realized using elements such as gates that are in turn probabilistic, are shown to form a natural basis for primitives in the signal processing (DSP) domain. In this paper, we show that probabilistic arithmetic can be used to compute the FFT in an extremely energy-efficient manner, yielding energy savings of over 5. 6X in the context of the widely used synthetic aperture radar (SAR) application [1]. Our results are derived using novel probabilistic cmos (PC-MOS) technology, characterized and applied in the past to realize ultra-efficient architectures for probabilistic applications [2, 3, 4]. When applied to the dsp domain, the resulting error in the output of a probabilistic arithmetic primitive, such as an adder for example, manifests as degradation in the signal-to-noise ratio (SNR) ofthe sar image that is reconstructed through the FFT algorithm. In return for this degradation that is enabled by our probabilistic arithmetic primitives ?- degradation visually indistinguishable from an image reconstructed using conventional deterministic approaches -- significant energy savings and performance gains are shown to be possible per unit of SNR degradation. These savings stem from a novel method of voltage scaling, which we refer to as biased voltage scaling (or BIVOS), that is the major technical innovation on which our probabilistic designs are based.

Major impediments to technology scaling in the nanometer regime include power (or energy) dissipation and “erroneous” behavior induced by process variations and noise susceptibility. In this paper, we demonstrate that CMOS devices whose behavior is rendered probabilistic by noise (yielding probabilistic CMOS or PCMOS) can be harnessed for ultra low energy and high performance computation. PCMOS devices are inherently probabilistic in that they are guaranteed to compute correctly with a probability 1= 2< p< 1 and thus, by design, they are expected to compute incorrectly with a probability (1 p). In this paper, we show that PCMOS technology yields significant improvements, both in the energy consumed as well as in the performance, for probabilistic applications with broad utility. These benefits are derived using an application-architecture-technology (A2T) co-design methodology introduced here, yielding an entirely novel family of probabilistic system-on-a-chip (PSOC) architectures. All of our application and architectural savings are quantified using the product of the energy and the performance denoted (energy×performance): the PCMOS based gains are as high as a substantial multiplicative factor of over 560 when compared to a competing energy-efficient CMOS based realization.

Highly scaled CMOS devices in the nanoscale regime would inevitably exhibit statistical or probabilistic behavior. Such behavior is due to process variations and other perturbations such as noise. Therefore current circuit design methodologies, which depend on the existence of deterministic and uniform devices with no consideration for either power consumption or probabilistic behavior, would no longer be sufficient to design robust circuits. To help overcome this challenge, we have been characterizing CMOS devices with probabilistic behavior (probabilistic CMOS or PCMOS devices) at several levels: from foundational principles to analytical modeling, simulation, fabrication and measurement, as well as innovative approaches to harnessing PCMOS devices in system-on-a-chip architectures which can implement a wide range of applications. In this paper, we present a broad overview of our contributions in the domain of PCMOS, and outline ongoing work and future challenges in this area.

Previous work has shown that a system-on-a-chip lock cache (SoCLC) reduces on-chip memory traffic, provides a fair and fast lock hand-off, simplifies software, increases the real-time predictability of the system and improves performance. In this research work, we extend the SoCLC mechanism with a priority inheritance support implemented in hardware. Priority inheritance provides a higher level of real-time guarantees for synchronizing application tasks. Experimental results indicate that our SoCLC hardware mechanism with priority inheritance achieves a 36% speedup in lock delay, 88% speedup in lock latency, and 15% speedup in the overall execution time when compared to its software counterpart. The cost in terms of additional hardware area for the SoCLC with priority inheritance is approximately 10000 NAND2 gates.

Intertask/interprocess synchronization overheads may be significant in a multiprocessor-shared memory System-on-a-Chip implementation. These overheads are observed in terms of lock latency, lock delay and memory bandwidth consumption in the system. It has been shown that a hardware solution brings a much better performance improvement than the synchronization algorithms developed in software [3]. Our previous work presented a SoC Lock Cache (SoCLC) hardware mechanism which resolves the Critical Section (CS) interactions among multiple processors and improves the performance criteria in terms of lock latency, lock delay and bandwidth consumption in a shared memory multi-processor SoC for short CSes [1]. This paper extends our previous work to support long CSes as well. This combined support involves modifications both in the RTOS kernel level facilities (such as support for preemptive versus non-preemptive synchronization, interrupt handling and RTOS initialization) and in the hardware mechanism. The worst-case simulation results of a database application model with client-server pair of tasks on a four-processor system showed that our mechanism achieved a 57% improvement in lock latency, 14% speed up in lock delay and a 35% overall speedup in total execution time.

For scalable-shared memory multiprocessor System-on-a-Chip implementations, synchronization overhead may cause catastrophic stalls in the system. Efficient improvements in the synchronization overhead in terms of latency, memory bandwidth, delay and scalability of the system involve a solution in hardware rather than in software. This paper presents a novel, efficient, small and very simple hardware unit that brings significant improvements in all of the above criteria: in an example, we reduce time spent for lock latency by a factor of 4.8, the worst-case execution of lock delay in a database application by a factor of more than 450. Furthermore, we developed a software architecture together with RTOS support to leverage our hardware mechanism. The worst-case simulation results of a client-server example on a four-processor system showed that our mechanism achieved an overall speedup of 27%.