-
MORCIC: Model Order Reduction Techniques for Electromagnetic Models of Integrated Circuits
Authors:
Dimitrios Garyfallou,
Athanasios Stefanou,
Christos Giamouzis,
Moschos Antoniadis,
Georgios Chararas,
Konstantinos Chatzis,
Dimitris Samaras,
Rafaela Themeli,
Anastasios Michailidis,
Vasiliki Gogolou,
Nikos Zachos,
Nestor Evmorfopoulos,
Thomas Noulis,
Vasilis F. Pavlidis,
Alkiviadis Hatzopoulos,
Elpida Chatzineofytou,
Yiannis Moisiadis
Abstract:
Model order reduction (MOR) is crucial for the design process of integrated circuits. Specifically, the vast amount of passive RLCk elements in electromagnetic models extracted from physical layouts exacerbates the extraction time, the storage requirements, and, most critically, the post-layout simulation time of the analyzed circuits. The MORCIC project aims to overcome this problem by proposing…
▽ More
Model order reduction (MOR) is crucial for the design process of integrated circuits. Specifically, the vast amount of passive RLCk elements in electromagnetic models extracted from physical layouts exacerbates the extraction time, the storage requirements, and, most critically, the post-layout simulation time of the analyzed circuits. The MORCIC project aims to overcome this problem by proposing new MOR techniques that perform better than commercial tools. Experimental evaluation on several analog and mixed-signal circuits with millions of elements indicates that the proposed methods lead to x5.5 smaller ROMs while maintaining similar accuracy compared to golden ROMs provided by ANSYS RaptorX.
△ Less
Submitted 14 November, 2023;
originally announced January 2024.
-
A New Dataflow Implementation to Improve Energy Efficiency of Monolithic 3D Systolic Arrays
Authors:
Prachi Shukla,
Vasilis F. Pavlidis,
Emre Salman,
Ayse K. Coskun
Abstract:
Systolic arrays are popular for executing deep neural networks (DNNs) at the edge. Low latency and energy efficiency are key requirements in edge devices such as drones and autonomous vehicles. Monolithic 3D (MONO3D) is an emerging 3D integration technique that offers ultra-high bandwidth among processing and memory elements with a negligible area overhead. Such high bandwidth can help meet the ev…
▽ More
Systolic arrays are popular for executing deep neural networks (DNNs) at the edge. Low latency and energy efficiency are key requirements in edge devices such as drones and autonomous vehicles. Monolithic 3D (MONO3D) is an emerging 3D integration technique that offers ultra-high bandwidth among processing and memory elements with a negligible area overhead. Such high bandwidth can help meet the ever-growing latency and energy efficiency demands for DNNs. This paper presents a novel implementation for weight stationary (WS) dataflow in MONO3D systolic arrays, called WS-MONO3D. WS-MONO3D utilizes multiple resistive RAM layers and SRAM with high-density vertical interconnects to multicast inputs and perform high-bandwidth weight pre-loading while maintaining the same order of multiply-and-accumulate operations as in native WS dataflow. Consequently, WS-MONO3D eliminates input and weight forwarding cycles and, thus, provides up to 40% improvement in energy-delay-product (EDP) over the native WS implementation in 2D at iso-configuration. WS-MONO3D also provides 10X improvement in inference per second per watt per footprint due to multiple vertical tiers. Finally, we also show that temperature impacts the energy efficiency benefits in WS-MONO3D.
△ Less
Submitted 7 January, 2024;
originally announced January 2024.
-
Temperature-Aware Monolithic 3D DNN Accelerators for Biomedical Applications
Authors:
Prachi Shukla,
Vasilis F. Pavlidis,
Emre Salman,
Ayse K. Coskun
Abstract:
In this paper, we focus on temperature-aware Monolithic 3D (Mono3D) deep neural network (DNN) inference accelerators for biomedical applications. We develop an optimizer that tunes aspect ratios and footprint of the accelerator under user-defined performance and thermal constraints, and generates near-optimal configurations. Using the proposed Mono3D optimizer, we demonstrate up to 61% improvement…
▽ More
In this paper, we focus on temperature-aware Monolithic 3D (Mono3D) deep neural network (DNN) inference accelerators for biomedical applications. We develop an optimizer that tunes aspect ratios and footprint of the accelerator under user-defined performance and thermal constraints, and generates near-optimal configurations. Using the proposed Mono3D optimizer, we demonstrate up to 61% improvement in energy efficiency for biomedical applications over a performance-optimized accelerator.
△ Less
Submitted 29 March, 2022;
originally announced March 2022.
