Accurate characterization of radiation-induced soft errors is a critical step toward understanding the impact of these glitches on circuit and system reliability. With process scaling, there has been exponential increase in number of transistors that can be packed on a die which, in turn, results in higher sensitive node count and persistent soft error susceptibilities. In this work, a novel circuit technique employing higher sensitivity toward soft errors is proposed. The circuit makes use of current-starved gates with bias knobs to fine-tune both measurement resolution and strike sensitivity enabling accelerated and efficient induction of errors in a limited-time irradiation test environment. The back-sampling chain (BSC) circuit can measure individual radiation-induced transient pulse with as low amplitude as$0.3\times $VDD while maintaining a high measurement resolution for pulsewidth characterization. The bias knobs allowing tuning of sensitivity and resolution enable, for the first time, a strike pulse waveform reconstruction methodology that can be used to calibrate current pulse models for assessing soft error rate (SER) sensitivity of standard logic gates.

Using a neural sampling approach, networks of stochastic spiking neurons, interconnected with plastic synapses, have been used to construct computational machines such as Restricted Boltzmann Machines (RBMs). Previous work towards building such networks achieved lower performances than traditional RBMs. More recently, Synaptic Sampling Machines (SSMs) were shown to outperform equivalent RBMs. In Synaptic Sampling Machines (SSMs), the stochasticity for the sampling is generated at the synapse. Stochastic synapses play the dual role of a regularizer during learning and an efficient mechanism for implementing stochasticity in neural networks over a wide dynamic range. In this paper we show that SSMs with stochastic synapses implemented in FPGA-based spiking neural networks can obtain a high accuracy in classifying MNIST handwritten digit database. We compare classification accuracy for different bit precision for stochastic and non-stochastic synapses and further argue that stochastic synapses have the same effect as synapses with higher bit precision but require significantly lower computational resources.

This session brings together specialists from the DfT, DfY and DfR domains that will address key problems together with their solutions for the 14 nm node and beyond, dealing with extremely complex chips affected by high defect levels, unpredictable and heterogeneous timing behavior, circuit degradation over time, including extreme situations related with the ultimate CMOS nodes, where all processor nodes, routers and links of single-chip massively parallel tera-device processors could comprise timing faults (such as delay faults or clock skews); a large percentage of these parts are affected by catastrophic failures; all parts experience significant performance degradations over time; and new catastrophic failures occur at low MTBF.

Traditional processors are designed to guarantee error-free operation under worst-case (1) device & interconnect parameter variations resulting from less than ideal manufacturing process control; (2) static & erratic defects; (3) operating environments such as temperature excursions and voltage droops; (4) critical path activation and path delay degradations due to multiple inputs switching simultaneously in gates containing transistor stacks, or signal coupling from neighboring lines in interconnect paths; (5) speed degradation over the operating lifetime due to transistor aging under voltage, temperature & current stress; (6) early-life failures due to latent defect accelerations; and (7) soft error due to cosmic rays and alpha particle impacts. The voltage-frequency settings for all processors are set based on these infrequently encountered worst-case considerations, even though under typical conditions voltage can be pushed down further or frequency increased without causing errors for most of the processors, thus limiting both energy efficiency and performance in scaled CMOS technologies.

Conventional microprocessors require a clock frequency (F CLK ) guardband to ensure correct functionality during infrequent dynamic operating variations in supply voltage (V CC ), temperature, and transistor aging. Consequently, these inflexible designs cannot exploit opportunities for higher performance by increasing F CLK or lower energy by reducing V CC during favorable operating conditions. This presentation describes a 45nm resilient microprocessor with error-detection and recovery circuits to detect and correct timing errors from dynamic variations to mitigate the F CLK guardband, thus enabling higher performance or lower energy as compared to a conventional design. The microprocessor core supports two distinct error-detection designs and two separate error-recovery techniques, allowing a direct comparison of the relative trade-offs. Silicon measurements demonstrate that resilient circuits enable a 41% throughput gain at equal energy or a 22% energy reduction at equal throughput, as compared to a conventional design when executing a benchmark program with a 10% V CC droop. In addition, the resilient circuits guide an adaptive clock controller that tracks recovery cycles and adapts to persistent variations by changing F CLK . The combination of error-detection and recovery circuits with dynamic adaptation allows the microprocessor to adapt to the operating environment to deliver maximum efficiency. The presentation concludes by discussing the opportunity of applying resilient techniques to enhance the dynamic operating range (i.e., high-performance and low-power modes) for microprocessors.

One of the most effective techniques to reduce a processor's power consumption is to reduce supply voltage. However, reducing voltage in the context of manufacturing-induced parameter variations can cause many types of memory circuits to fail. As a result, voltage scaling is limited by a minimum voltage, often called Vccmin, beyond which circuits may not operate reliably. Large memory structures (e.g., caches) typically set Vccmin for the whole processor. In this paper, we propose two architectural techniques that enable microprocessor caches (L1 and L2), to operate at low voltages despite very high memory cell failure rates. The Word-disable scheme combines two consecutive cache lines, to form a single cache line where only non-failing words are used. The Bit-fix scheme uses a quarter of the ways in a cache set to store positions and fix bits for failing bits in other ways of the set. During high voltage operation, both schemes allow use of the entire cache. During low voltage operation, they sacrifice cache capacity by 50% and 25%, respectively, to reduce Vccmin below 500mV. Compared to current designs with a Vccmin of 825 mV, our schemes enable a 40% voltage reduction, which reduces power by 85% and energy per instruction (EPI) by 53%.

In this paper, an accurate approach for estimating SRAM dynamic stability is proposed. The conventional methods of SRAM stability estimation suffer from two major drawbacks: 1) using static failure criteria, such as static noise margin (SNM), which does not capture the transient and dynamic behavior of SRAM operation and 2) using quasi-Monte Carlo simulation, which approximates the failure distribution, resulting in large errors at the tails where the desired failure probabilities exist. These drawbacks are eliminated by employing a new distribution-independent, most-probable-failure-point search technique for accurate probability calculation along with accurate simulation-based dynamic failure criteria. Compared to previously published techniques, the proposed technique offers orders of magnitude improvement in accuracy. Furthermore, the proposed technique enables the correct evaluation of stability in real operation conditions and for different dynamic circuit techniques, such as dynamic write-back, where the conventional methods are not applicable.

We present a new methodology which takes into consideration the effect of within-die (WID) process variations on a low-voltage parallel system. We show that in the presence of process variations one should use a higher supply voltage than would otherwise be predicted to minimize the power consumption of a parallel systems. Previous analyses, which ignored WID process variations, provide a lower nonoptimal supply voltage which can underestimate the energy/operation by 8.2. We also present a novel technique to limit the effect of temperature variations in a parallel system. As temperatures increases, the scheme reduces the power increase by 43% allowing the system to remain at it's optimal supply voltage across different temperatures. To further limit the effect of variations, and allow for a reduced power consumption, we analyzed the effects of clustering. It was shown that providing different voltages to each cluster can provide a further 10% reduction in energy/operation to a low-voltage parallel system, and that the savings by clustering increase as technology scales.

On-chip serial link buses have been previously proposed as a strong solution to reduce the complexity and/or the energy dissipation of on-chip interconnect fabrics. However, it was noticed that serializing m-bits on a single interconnect (serial-link) increases the overall data switching activity. This paper presents a quantitative analysis of the switching activity of serial links, and provides closed form expressions for the average activity factors. Two transition encoding schemes, to reduce the activity factor of serial links, are discussed and analyzed. The impact of the encoding schemes on the MCF between neighboring interconnects is also discussed. The analysis shows that both of the schemes provide significant reduction in the average activity factor and energy dissipation reduction, but each in a different range of input activity factors. The two encoding bus schemes were modeled in a 70nm CMOS technology, and compared to an unencoded serial link bus and a parallel line bus. Simulation results show that the transition encoded bus schemes reduce the overall energy dissipation of the unencoded serial link bus by up to 96%

SRAM arrays using differential sensing (DS) and single-ended sensing (SE) are designed and fabricated in a test chip and their power and performance behaviors are studied in this paper. Sense amplifier offset (DC condition), which is one of the main criterion to determine the required bit-line differential, is measured. A novel SE scheme is proposed to overcome the delay degradation due to large bit line leakage in scaled technology. With marginal switching power savings, the SE array is 56% slower than the DS array in 90 nm technology with a single high-Vt at 350 mV. The difference narrows down to 30% in low-Vt case. Using asymmetric cells, instead of symmetric cells, in single-ended large signal arrays improves delay by 3%, while the power consumption remains approximately the same.

Insertion of time-borrowing (TB) flip-flops in multi-cycle repeater-based on-chip interconnects enables significant improvements in mean performance and energy by averaging systematic and random within-die (WID) delay variations across multiple interconnect segments. A statistically-based analytical model is derived to design a TB N-cycle interconnect with optimal delay variation tolerance. The model elucidates the dependency of the transparency window required to achieve data delay averaging on the delay variation mismatch between interconnect segments. Statistical circuit simulations and analyses in a 65nm process technology demonstrate that TB multi-cycle interconnects enable a 4-6% mean maximum clock frequency (FMAX) improvement and a corresponding 10% average energy savings over optimally designed multi-cycle interconnects with conventional master-slave flip-flops. The maximum mean FMAX benefit ranges from 4.0-7.5%, corresponding to approximately a bin-split shift in the FMAX distribution. For 1.41X larger WID delay variations, the maximum mean FMAX gain rises to 5-10%.

Passive shielding has been used to reduce the capacitive coupling effects of adjacent bus lines by inserting passive ground or power lines (shields) between them. Active shielding is another shielding technique in which the shield is allowed to switch depending on the switching pattern of its adjacent bus lines. This paper formally derives the optimal active shielding logic function for minimum power dissipation. It is also shown that this optimal active shielding architecture depends on the ratio of coupling to ground capacitance (/spl gamma/=C/sub c//C/sub g/). Optimal active shielding is shown to provide up to 25% reduction in bus power dissipation compared to conventional passive shielding. A suboptimal active shielding architecture with simpler hardware is also proposed. Theoretically, using the suboptimal shielding architecture leads to less than 6% bus power penalty compared to the optimal active shielding logic circuit. However, due to the simpler shield encoding circuitry, simulation results show that the suboptimal active shielding architecture leads to higher overall energy savings compared to the optimal active shielding architectures.

We present a new methodology which takes into consideration the effect of Within-Die (WID) process variations on a low-voltage parallel system. We show that in the presence of process variations one should use a higher supply voltage than would otherwise be predicted to minimize the power consumption of a parallel systems. Previous analyses, which ignored WID process variations, provide a lower non-optimal supply voltage which can underestimate the energy/ operation by 8.2X. We also present a novel technique to limit the effect of temperature variations in a parallel system. As temperatures increases, the scheme reduces the power increase by 43% allowing the system to remain at it's optimal supply voltage across different temperatures.

As technology scales, the shrinking wire width increases the interconnect resistivity, while the decreasing interconnect spacing significantly increases the coupling capacitance. This paper proposes reducing the number of bus lines of the conventional parallel-line bus CB architecture by multiplexing each m-bits onto a single line. This bus architecture, the serial-link bus SLB, transforms an n-bit conventional parallel-line bus into an n/m-line (serial-link) bus. The advantage of serial-link buses is that they have fewer lines, and if the bus width is kept the same, serial- link buses will have larger line width and spacing. Increasing the line width has a twofold reduction effect on the line resistance, as the resistivity of sub-100 nm wires significantly drops as the line width increases. Also, increasing the line width and spacing reduces the coupling capacitance between adjacent lines, but increases the line-to-ground capacitance. Thus, an optimum degree of multiplexing m exists that minimizes the bus energy dissipation and maximizes the bus throughput per-unit area. The optimum degree of multiplexing for maximum throughput-per- unit-area and for minimum energy dissipation for the 25-130 nm technologies was determined in this paper. HSPICE simulations show that; for the same throughput-per-unit-area as conventional parallel-line buses, the serial-link bus architecture reduces the energy dissipation by up to 31.42% for a 64-bit bus implemented in an intermediate metal layer of a 50 nm technology and a reduction of 52.7% is projected for the 25 nm technology.

This paper proposes a bus architecture called skewed repeater bus (SRB) for reducing on-chip interconnect energy in microprocessors. By introducing relative delay between neighboring bus lines, SRB reduces both average and worst-case coupling capacitance between those lines. SRB is compared to previously published techniques like delayed data bus (DDB) and delayed clock bus (DCB). Simulation results in 65-nm process show that bus energy reduction of 18% is achieved when SRB is applied to a real microprocessor example, versus 11% and 7% only for DDB and DCB; respectively.

This paper presents a new approach for energy reduction and speed improvement of multiport SRAMs. The key idea is to use current-mode for both read and write operations. To toggle a memory cell, a very small voltage swing is first created on the high-capacitive bit lines. This voltage is then translated into a differential current being injected into the cell, which in turn allows complementary potential to be developed on the cell nodes. As compared to the conventional write approach, SPICE simulations using a 0.35-/spl mu/m CMOS process have shown 2.8 to 9.9/spl times/ in energy savings and 1.02 to 6.36/spl times/ reduction in delay, for memory sizes of 32 to 1 K words. We also present a current-mode sense-amplifier that operates in a similar fashion as the write circuit. The design and implementation of a pipelined 32/spl times/64 three-port register file utilizing the proposed technique is described. Measurements of the register file chip have confirmed the feasibility of the approach.

A new approach to modeling the decoding hierarchy in a hierarchical word line (HWL) SRAM architecture using integer-linear programming (ILP) is introduced. Using this approach, the HWL architecture is shown to be inadequate for very large SRAM sizes. Alternatively, a new low-power high-speed SRAM architecture is described. This architecture is shown to have fairly constant speed and power dissipation for sizes ranging between 32 kb to 4 Mb. Low-power is achieved by a voltage boosting technique not requiring a two-step voltage and by a new method of tristating memory cells during a write operation. The SRAM was implemented in a 0.35 /spl mu/m CMOS technology operated at 150 MHz while dissipating only 10 mW.

This paper presents a simple, yet efficient method to characterize the effective capacitance in data-path macros for architectural-level power estimation. Given a library of hard-macros, a capacitance model based on linear regression is derived for each macro. A transistor-level tool is employed for capacitance extraction. The capacitance models can be used during architectural-level power estimation. Unlike previous approaches, our characterization methodology assumes no specific word-level statistics of the input data, requires little knowledge about the structure of the modules, allows the user to trade-off accuracy and characterization time, and propagates effective capacitance directly from transistor-level (real) implementations. Simulation experiments on a set of data-path components with various sizes are performed. Compared to a previously published approach, our scheme significantly improves the accuracy of RTL power estimation and produces results within 15% from a transistor-level tool on the average.