Bridging the gap between heterogeneous computing and next generation memory architecture using high level synthesis

1 online resource (102 pages) : PDF

2019

University of North Carolina at Charlotte

The impact of transistor scaling on FPGAs is changing the role of FPGA from accelerators to a major role as processors. With this rapid development and the ability to implement complex systems on FPGAs, the conventional hardware language design flow is making way for software-like language using High-Level Synthesis (HLS). While academic and commercial HLS tools have made huge strides, nearly all these tools focus exclusively on the computation and the data path. Rarely do they directly address the memory subsystem and its impact onthe overall performance. At best, the programmers can assist thetools with optimization which indirectly impact the memory subsystem performance. This has (unintentionally) exacerbated the already existing memory issues. The performance of DDR memory which has been the main stream off-chip memory has been lagging behind the processor performance. This has resulted in a performance gap and emerging memories such as Hybrid Memory Cube (HMC), High Bandwidth \Memory (HBM), and others are promising prospects in reducing this gap. However, integrating these new memory technologies with HLS design flow has not been trivial. To fully utilize the performance benefits, the programmer must understand the low-level details of the hardware.In this work, we conduct a systematic analysis of different HLS generated circuits on different off-chip memories. Our analysis identifies the root cause of the problem which is how the low-level hardware interacts with the memory subsystem.To mitigate these issues we introduce a hardware middle layer which establishes compatibility and a software transformation to improve the performance. This design is compared with the baseline system to evaluate the performance improvement from the methodology on heterogeneous systems.

doctoral dissertations

Electrical engineering
Computer engineering

Ph.D.

FPGA
HLS
Next Generation Memory

Electrical Engineering

Sass, Ronald

Conrad, James
Weldon, Thomas
Schmidt, Andrew
Saule, Erik

Thesis (Ph.D.)--University of North Carolina at Charlotte, 2019.

This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s). For additional information, see http://rightsstatements.org/page/InC/1.0/.

Copyright is held by the author unless otherwise indicated.

Rajagopala_uncc_0694D_12257

http://hdl.handle.net/20.500.13093/etd:2390