Nanoscale Computing Fabrics Lab - Nanoarchitecture/Devices/Circuits

SkyBridge-3D-CMOS 2.0: IC Technology for Stacked-Transistor 3D ICs beyond FinFETs

Sourabh Kulkarni — Sat, 17 Jul 2021 16:31:26 +0000

Sachin Bhat

Mingyu Li

Sounak 'Shaun' Ghosh

Sourabh Kulkarni

Csaba Andras Moritz

2021

2021 ISVLSI Nagarajan Ranganathan Best Paper Award

Abstract:

For sub-5nm technology nodes, gate-all-around (GAA) FETs are positioned to replace FinFETs to enable the continued miniaturization of ICs in the future. In this paper, we introduce SkyBridge-3D-CMOS 2.0, a 3D-IC technology featuring integration of stacked vertical GAAFETs and 3D interconnects. It aims to provide an integrated solution to critical technology aspects, especially when scaling to sub-5nm nodes. We address important aspects such as 3D fabric components, CAD tool flow, compact model for the GAAFETs and a scalable manufacturing process. The fabric features junctionless accumulation-mode field effect transistors (JAMFETs) including various configurations with multiple threshold voltages and multiple nanowires per transistor, to meet performance and stand-by power constraints of modern SoCs. Furthermore, we develop BSIM-CMG-based compact models for these device configurations to enable technology assessment using SPICE simulations. To enable scalable manufacturing, we create virtual process decks incorporating etch and deposition models using Process Explorer, an industry standard process emulation tool. Technology assessment using ring oscillators shows that SkyBridge-3D-CMOS 2.0 at the chosen design point, using 16nm gate length and 10-nm nanowires, achieves ∼18% performance and 31% energy efficiency improvement versus 7nm FinFET CMOS. Area analysis of standard cells shows up to 6x benefit versus aggressively scaled 2D-5T cells.

PDF

Publication Type:

Conference

Vol. No.:

Issue No.:

Article No.:

citation:

IEEE Computer Society Annual Symposium on VLSI

Architecting for Artificial Intelligence with Emerging Nanotechnology

Sourabh Kulkarni — Sun, 31 Jan 2021 15:39:03 +0000

Sourabh Kulkarni

Sachin Bhat

Csaba Andras Moritz

2021

Abstract:

Artificial Intelligence is becoming ubiquitous in products and services that we use daily. Although the domain of AI has seen substantial improvements over recent years, its effectiveness is limited by the capabilities of current computing technology. Recently, there have been several architectural innovations for AI using emerging nanotechnology. These architectures implement mathematical computations of AI with circuits that utilize physical behavior of nanodevices purpose-built for such computations. This approach leads to a much greater efficiency vs. software algorithms running on von-Neumann processors or CMOS architectures which emulate the operations with transistor circuits. In this paper, we provide a comprehensive survey of these architectural directions and categorize them based on their contributions. Furthermore, we discuss the potential offered by these directions with real world examples. We also discuss major challenges and opportunities in this field.

Authors Sourabh Kulkarni and Sachin Bhat contributed equally to this work.

PDF (In Press)

Publication Type:

Conference

Vol. No.:

Issue No.:

Article No.:

citation:

ACM Journal on Emerging Technologies in Computing Systems (JETC)

A Wafer-scale Manufacturing Pathway for Fine-grained Vertical 3D-IC Technology

Sourabh Kulkarni — Mon, 25 Jan 2021 21:04:37 +0000

2021

Abstract:

Three-dimensional integrated circuits (3D-ICs) provide a feasible path for scaling CMOS technology in the foreseeable future. IMEC and IRDS roadmaps project that 3D integration is a key avenue for the IC industry beyond 2024. They project that some form of 3D-IC technology based on nanosheets/nanowires is likely to become mainstream soon. SkyBridge-3D-CMOS (S3DC) is one among the first vertical nanowire-based fine-grained 3D-IC directions which offers paradigm shift in technology scaling as well as design. Rather than die-die and layer-layer stacking, S3DC’s core aspects, from device to circuit style to interconnect, are co-architected in a 3D fabric-centered manner building on a uniform 3D nanowire template. Nanowire-based 3D-IC technologies such as S3DC solve most of the traditional scaling issues of 2D-CMOS but present new manufacturing challenges because of their complex 3D geometry. Therefore, for these directions to become mainstream, a robust wafer-scale manufacturing pathway that addresses these challenges is vital. In this paper, we propose a wafer-scale manufacturing pathway aimed at developing and optimizing the manufacturing process flows of S3DC. Using physics-driven virtual process integration functionalized with design and process parameters, we obtained realistic 3D structures for all the underlying IC elements and finally combined them to build3D standard cells in S3DC. Electrical characterization of resultant structures using process and device simulations were performed while considering the material properties and nanoscale physics effects. Circuit-level simulations accounting for device behavior using SPICE-compatible compact model and circuit interconnect parasitics were carried out to study the impact of variations in process steps such as patterning, lithography, etch, deposition on device and interconnect performance. Our bottom-up simulation results indicate that the proposed pathway is robust enough to be adopted for large-scale production thus paving the way for wide-spread adoption of vertical fine-grained 3D-IC technologies.

PDF

Publication Type:

Conference

Vol. No.:

Issue No.:

Article No.:

citation:

in Proceedings of International Symposium on Quality Electronic Design (ISQED)

Circuit Design Steps for Nano-Crossbar Arrays: Area-Delay-Power Optimization with Fault Tolerance

Sourabh Kulkarni — Tue, 15 Oct 2019 01:58:15 +0000

2019

Abstract:

Nano-crossbar arrays have emerged to achieve high performance computing beyond the limits of current CMOS. They offer area and power efficiency in courtesy of their easyto-fabricate and dense physical
structures consisting of regularly placed crosspoints as computing elements. Depending on the used technology, a crosspoint behaves as a diode, a memristor, a field effect transistor, or a four-terminal switching device. In this study, we comparatively elaborate on these technologies in terms of their capabilities for computing in terms of area, delay, and power consumption. Also, we consider fault tolerance capabilities of the arrays. Due to the stochastic nature of nano-fabrication, nanoarrays have much higher fault rates compared conventional technologies such as CMOS. As a result, this study introduces a synthesis methodology that considers basic technology preference for switching crosspoints and defect or fault rates of the given nanoarray as well as their effects on performance metrics including power, delay, and area.

PDF

Publication Type:

Journal

Vol. No.:

Issue No.:

pages:

1-11

Article No.:

citation:

IEEE Transactions on Nanotechnology

SkyNet: Memristor-based 3D IC for Artificial Neural Networks

Sourabh Kulkarni — Tue, 26 Sep 2017 15:18:27 +0000

2017

Abstract:

Hardware implementations of artificial neural networks (ANNs) have become feasible due to the advent of persistent 2-terminal devices such as memristor, phase change memory, MTJs, etc. Hybrid memristor crossbar/CMOS systems have been studied extensively and demonstrated experimentally. In these circuits, memristors located at each cross point in a crossbar are, however, stacked on top of CMOS circuits using back end of line processing (BOEL), limiting scaling. Each neuron’s functionality is spread across layers of CMOS and memristor crossbar and thus cannot support the required connectivity to implement large-scale multi-layered ANNs. This paper introduces a new fine-grained 3D integrated ASIC technology for ANNs that is the first IC technology for this purpose. Synaptic weights implemented with devices are incorporated in a uniform vertical nanowire template co-locating the memory and computation requirements of ANNs within each neuron. Novel 3D routing features are used for interconnections in all three dimensions between the devices enabling high connectivity without the need for special pins or metal vias. To demonstrate the proof of concept of this fabric, classification of binary images using a perceptron-based feed forward neural network is shown. Bottom-up evaluations for the proposed fabric considering 3D implementations of fabric components reveal up to 21x density, 1.8x power benefits and a 2.6x improvement in delay when compared to 16nm hybrid memristor/CMOS technology.

PDF

Publication Type:

Conference

Vol. No.:

Issue No.:

pages:

In Press

Article No.:

citation:

in Proceedings of IEEE/ACM International Symposium on Nanoscale Architectures (NanoArch)

Fine-Grained 3D Reconfigurable Computing Fabric with RRAM

Mingyu Li — Wed, 20 Sep 2017 13:56:20 +0000

2017

Abstract:

Non-volatile 3D FPGA research to date utilizes layer-by-layer stacking of 2D CMOS / RRAM circuits. On the other hand, vertically-composed 3D FPGA that integrates CMOS and RRAM circuits has eluded us, owing to the difficult requirement of highly customized regional doping and material insertion in 3D to build and route complementary p- and n-type transistors as well as resistive switches. In the layer-by-layer nonvolatile 3D FPGA, the connectivity between the monolithically stacked RRAMs and underlying CMOS circuits is likely to be limited and lead to large parasitic RCs. In this paper, we propose a fine-grained 3D reconfigurable computing fabric concept. It implements CMOS / RRAM hybrid circuits within the pre-doped vertical nanowire template. Transistors and resistive switches can be integrated with a fine granularity, which reduces the routing overhead between RRAM and CMOS circuits and increases the density. We estimate the density benefit of the proposed fabric to be 27X relative to the monolithic 3D FPGA with stacked RRAMs. Estimated Elmore delays are improved by 5.4X and 2.2X for configuration and normal operation, respectively.

PDF

Publication Type:

Conference

Vol. No.:

Issue No.:

pages:

in press

Article No.:

citation:

in Proceedings of IEEE/ACM International Symposium on Nanoscale Architectures (NanoArch)

A 14nm FinFET transistor-level 3D partitioning design to enable high-performance and low-cost monolithic 3D IC

JiaJun Shi — Tue, 19 Sep 2017 20:08:20 +0000

2016

Abstract:

Monolithic 3D IC (M3D) shows degradation in performance compared to 2D IC due to the restricted thermal budget during fabrication of sequential device layers. A transistor-level (TR-L) partitioning design is used in M3D to mitigate this degradation. Silicon validated 14nm FinFET data and models are used in a device-to-system evaluation to compare the TR-L partitioned M3D’s (TR-L M3D) performance against the conventional gate-level (G-L) partitioned M3D’s performance as well as standard 2D IC. Extensive cell-level and system-level evaluation, including various device and interconnect process options, shows that the TR-L M3D provides up to 20% improved performance while still maintaining around 30% power saving compared to standard 2D IC. Additionally, the TR-L partitioning design enables M3D with a simplified process flow that leads to 23% lower cost compared to that of G-L partitioning scheme.

PDF

Publication Type:

Conference

Vol. No.:

Issue No.:

pages:

2.5.1 - 2.5.4

Article No.:

citation:

IEEE International Electron Devices Meeting (IEDM)

Structure Discovery for Gene Expression Networks with Emerging Stochastic Hardware

Sourabh Kulkarni — Fri, 08 Sep 2017 14:46:39 +0000

Sourabh Kulkarni

Sachin Bhat

Csaba Andras Moritz

2017

Abstract:

Gene Expression Networks (GENs) attempt to model how genetic information stored in the DNA (Genotype) results in the synthesis of proteins, and consequently, the physical traits of an organism (Phenotype). Deciphering GENs plays an important role in a wide range of applications from genetic studies of the origins of life to personalized healthcare. Probabilistic graphical models such as Bayesian Networks (BNs) are used to perform learning and inference of GENs from genetic data. Current techniques of generating BNs of GENs from data, which are mostly approximate in nature, involve searching and scoring of multiple probabilistic graphical structures. However, while search algorithms can be efficiently implemented in software, the same is not true for scoring. Scoring of probabilistic models with inherent parallelism is inefficient when performed sequentially over conventional architectures comprising of deterministic devices. In this paper, we introduce a new nanoscale hardware acceleration framework, enabling fast and efficient Bayesian inference operations, significantly accelerating the scoring aspect of the BN learning of GENs using a combination of emerging stochastic devices and CMOS technology. The stochasticity of the devices is utilized to efficiently perform approximate inference on probabilistic networks, and the circuit framework constituting these devices is designed to exploit the inherent parallelism in these models. We demonstrate approximate inference operation over a small BN. We estimate the performance benefits of five orders of magnitude in performing inference operations using this architecture over software-only approaches.

PDF

Publication Type:

Conference

Vol. No.:

Issue No.:

pages:

In Press

Article No.:

citation:

in Proceedings of IEEE International Conference on Rebooting Computing (ICRC)

Magneto-electric Approximate Computational Circuits for Bayesian Inference

Sourabh Kulkarni — Fri, 08 Sep 2017 14:43:58 +0000

2017

Abstract:

Probabilistic graphical models like Bayesian Networks (BNs) are powerful cognitive-computing formalisms, with many similarities to human cognition. These models have a multitude of real-world applications. New emerging-technology based circuit paradigms leveraging physical equivalence e.g., operating directly on probabilities vs. introducing layers of abstraction, have shown promise in raising the performance and overall efficiency of BNs, enabling networks with millions of random variables. While previous BNs of up to 100s of nodes have been shown to require single-digit precision without affecting application outcomes, the significantly larger number of variables requires the computational precision to be scaled to correctly support BN operations. We introduce a new computational circuit fabric based on mixed-signal magneto-electric computations operating with physical equivalence and supporting probabilistic computations with a new approximate circuit style. Precision scaling impacts area at a logarithmic vs. linear scale offering a much lower power and performance cost than in prior directions. Results show 30x area reduction for a 0.001 precision vs. prior direction, while maintaining three orders of magnitude benefits vs. 100-core processor implementations.

PDF

Publication Type:

Conference

Vol. No.:

Issue No.:

pages:

In press

Article No.:

citation:

in Proceedings of IEEE International Conference on Rebooting Computing (ICRC)

Skybridge-3D-CMOS: A Fine-Grained 3D CMOS Integrated Circuit Technology

Mingyu Li — Thu, 27 Apr 2017 14:51:17 +0000

2017

Abstract:

Parallel and monolithic 3D integration directions realize 3D integrated circuits (ICs) by utilizing layer-by-layer implementations, with each functional layer being composed in 2D. In contrast, vertically-composed 3D CMOS has eluded us likely due to the seemingly insurmountable requirement of highly customized complex routing and regional 3D doping to form and connect CMOS pull-up and pull-down networks in 3D. In the current layer-by-layer directions, routing can be worse than 2D CMOS because of the limited pin access. In this paper, we propose Skybridge-3D-CMOS (S3DC), an IC fabric that shows for the first time a pathway to achieve fine-grained static CMOS circuit implementations using the vertical direction while also solving 3D routability. It employs a new fabric assembly scheme based on pre-doped vertical nanowire bundles. It implements circuits in and across nanowires. It utilizes unique connectivity features to achieve CMOS connectivity in 3D with excellent routability. As compared to the usually severely congested monolithic 3D implementations, S3DC eliminates the routing congestions in all benchmarks studied. Further results, for the implemented benchmarks, show 56%-77% reductions in power consumption, 4X-90X increases in density, and 20% loss to 9% benefit in best operating frequencies compared with the transistor-level monolithic 3D technology.

PDF

Publication Type:

Journal

Vol. No.:

Issue No.:

pages:

639-652

Article No.:

citation:

IEEE Transactions on Nanotechnology, Special Issue on Revolutionary 3-D Integration