The present invention provides stretchable, and optionally printable, semiconductors and electronic circuits capable of providing good performance when stretched, compressed, flexed or otherwise deformed. Stretchable semiconductors and electronic circuits of the present invention preferred for some applications are flexible, in addition to being stretchable, and thus are capable of significant elongation, flexing, bending or other deformation along one or more axes. Further, stretchable semiconductors and electronic circuits of the present invention may be adapted to a wide range of device configurations to provide fully flexible electronic and optoelectronic devices.

http://science.sciencemag.org/content/sci/311/5758/208.full.pdf

Stretchable form of single crystal silicon for high performance electronics on rubber substrates

Regression Analysis of Count Data: Preface

For more than four decades, transistors have been shrinking exponentially in size, and therefore the number of transistors in a single microelectronic chip has been increasing exponentially. Such an increase in packing density was made possible by continually shrinking the metal–oxide–semiconductor field-effect transistor (MOSFET). In the current generation of transistors, the transistor dimensions have shrunk to such an extent that the electrical characteristics of the device can be markedly degraded, making it unlikely that the exponential decrease in transistor size can continue. Recently, however, a new generation of MOSFETs, called multigate transistors, has emerged, and this multigate geometry will allow the continuing enhancement of computer performance into the next decade.

Multigate transistors as the future of classical metal–oxide–semiconductor field-effect transistors

This paper describes an analytical model for the access and cycle times of on-chip direct-mapped and set-associative caches. The inputs to the model are the cache size, block size, and associativity, as well as array organization and process parameters. The model gives estimates that are within 6% of Hspice results for the circuits we have chosen. This model extends previous models and fixes many of their major shortcomings. New features include models for the tag array, comparator, and multiplexor drivers, nonstep stage input slopes, rectangular stacking of memory subarrays, a transistor-level decoder model, column-multiplexed bitlines controlled by an additional array organizational parameter, load-dependent size transistors for wordline drivers, and output of cycle times as well as access times. Software implementing the model is available via ftp.

CACTI: an enhanced cache access and cycle time model

Three-dimensional (3D) integrated circuits (ICs), which contain multiple layers of active devices, have the potential to dramatically enhance chip performance, functionality, and device packing density. They also provide for microchip architecture and may facilitate the integration of heterogeneous materials, devices, and signals. However, before these advantages can be realized, key technology challenges of 3D ICs must be addressed. More specifically, the processes required to build circuits with multiple layers of active devices must be compatible with current state-of-the-art silicon processing technology. These processes must also show manufacturability, i.e., reliability, good yield, maturity, and reasonable cost. To meet these requirements, IBM has introduced a scheme for building 3D ICs based on the layer transfer of functional circuits, and many process and design innovations have been implemented. This paper reviews the process steps and design aspects that were developed at IBM to enable the formation of stacked device layers. Details regarding an optimized layer transfer process are presented, including the descriptions of 1) a glass substrate process to enable through-wafer alignment; 2) oxide fusion bonding and wafer bow compensation methods for improved alignment tolerance during bonding; 3) and a single-damascene patterning and metallization method for the creation of high-aspect-ratio (6:1 108 vias/cm2), and extremely aggressive wafer-to-wafer alignment (submicron) capability.

Three-dimensional integrated circuits

RRAM-based in-memory computing (IMC) effectively accelerates deep neural networks (DNNs) and other machine learning algorithms. On the other hand, in the presence of RRAM device variations and lower precision, the mapping of DNNs to RRAM-based IMC suffers from severe accuracy loss. In this work, we propose a novel hybrid IMC architecture that integrates an RRAM-based IMC macro with a digital SRAM macro using a programmable shifter to compensate for the RRAM variations and recover the accuracy. The digital SRAM macro consists of a small SRAM memory array and an array of multiply-and-accumulate (MAC) units. The nonideal output from the RRAM macro, due to device and circuit nonidealities, is compensated by adding the precise output from the SRAM macro. In addition, the programmable shifter allows for different scales of compensation by shifting the SRAM macro output relative to the RRAM macro output. On the algorithm side, we develop a framework for the training of DNNs to support the hybrid IMC architecture through ensemble learning. The proposed framework performs quantization (weights and activations), pruning, RRAM IMC-aware training, and employs ensemble learning through different compensation scales by utilizing the programmable shifter. Finally, we design a silicon prototype of the proposed hybrid IMC architecture in the 65-nm SUNY process to demonstrate its efficacy. Experimental evaluation of the hybrid IMC architecture shows that the SRAM compensation allows for a realistic IMC architecture with multilevel RRAM cells (MLCs) even though they suffer from high variations. The hybrid IMC architecture achieves up to 21.9%, 12.65%, and 6.52% improvement in post-mapping accuracy over state-of-the-art techniques, at minimal overhead, for ResNet-20 on CIFAR-10, VGG-16 on CIFAR-10, and ResNet-18 on ImageNet, respectively.

Hybrid RRAM/SRAM in-Memory Computing for Robust DNN Acceleration

This paper describes power analysis at sub-zero temperatures for a high performance dynamic multiport register file (6 Read and 2 Write ports, 32 wordlines x 64 bitlines) fabricated in 0.25 μm Silicon on Insulator (SOI) and bulk technologies. Based on the hardware it is shown that the performance of both register file and latch improves by 2-3.5% per 10§ C reduction in temperature. The standby power for SOI reduces by 1.5% to 3 per 10§ C temperature drop down to -30§ C. The SOI chip is shown to have more significant performance improvement at low temperatures compared to bulk chip due to the floating body effect which partially offsets the increase in the threshold voltages (Vt). The low temperature performance gain is attributed to reduction in capacitance (around 7-8%) and rest is due to dynamic threshold voltages. At 30§ C the register file is capable of functioning close to 1.02 GHz for read and write operations in a single cycle.

"Cool low power" 1GHz multi-port register file and dynamic latch in 1.8 V, 0.25 μm SOI and bulk technology (poster session)

Nonvolatile latches are increasingly popular with the advent of IoT design. We study the yield energy tradeoff of the backup mechanism of an STT-MTJ based nonvolatile latch. For the yield analysis, we rely on Hicks and Wheeling methodology for multi-cone radial sampling for the purpose of rare fail estimation. Yield is shown to be delimited by the Parallel-to-AntiParallel magnetic angle transitions. To accommodate for the slower cells, we note an increase in the average energy requirements for the backup mechanism. Simulations indicate an increase of 10%–40% for the average energy requirements to achieve an ideal yield requirement close to 99% for different number of components.

Yield and energy tradeoffs of an NVLatch design using radial sampling

A method and a system including a processor performing a failure region exploration through uniform sampling of plurality of variables related to a circuit, the processor shifting probability distributions to explore failure probability, the processor estimating the failure probability and standard deviation by determining mean and standard deviation of failure probability of a circuit, the processor terminating sampling when a confidence interval bounds converge, and a peripheral device providing a report on the failure of the circuit when the sampling is terminated by the processor.

Massive multi-dimensionality failure analytics with smart converged bounds

The feasibility of nano-scale strained-Si technologies for low-power applications is studied. Static and dynamic power for strained-Si device is analyzed and compared with conventional bulk-Si technology. Optimum device design points are suggested, and strained-Si CMOS circuits are studied, showing substantially reduced power consumptions. The trade-offs for power and performance in strained-Si devices/circuits are discussed. Further, analysis and low-power design points are applied and extended to strained Si on SOI substrate (SSOI) CMOS technology.

Rajiv V. Joshi

Papers

Hybrid RRAM/SRAM in-Memory Computing for Robust DNN Acceleration

"Cool low power" 1GHz multi-port register file and dynamic latch in 1.8 V, 0.25 μm SOI and bulk technology (poster session)

Yield and energy tradeoffs of an NVLatch design using radial sampling

Massive multi-dimensionality failure analytics with smart converged bounds

Power analysis of strained-Si devices/circuits