{"title":"Intelligence Everywhere: The Challenges and Opportunities for Semiconductor Designs","authors":"Pin-Han Chen","doi":"10.1109/VLSI-DAT52063.2021.9427341","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427341","url":null,"abstract":"Technology across the industries is fascinated with the idea of Artificial Intelligence (AI). The worldwide market for AI accelerator chipsets is predicted to grow from $8 billion in 2019 to reach $70 billion by 2026 [1]. From automating repetitive tasks, recommending products and streamlining customer services improve business revenue and efficiency. The evolving neural network (NN) models running on CPUs, GPUs, FPGAs and ASICs have been exploited in various domains such as computer vision, natural language processing (NLP), and personal recommendation and successfully adapted in field applications. [2] Intelligence tends to be everywhere. From efficient point of view, ASICs are the most outstanding way. For edge AI devices, the total power consumption is in single to double digits mini-watt, and the size of chip is another winning point to make ASICs competitive. On the other hand, Cloud AI computing power is around dozen-Walt and takes more about total cost of ownership (TCO) [3]. In addition to product positioning differences, there are still major challenges for both edge and cloud AI/DL ASICs should deal with. In this paper, we’d like to discuss those challenges and also summarize relevant opportunities.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114931128","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A TinyMLaaS Ecosystem for Machine Learning in IoT: Overview and Research Challenges","authors":"Hiroshi Doyu, Roberto Morabito, Martina Brachmann","doi":"10.1109/VLSI-DAT52063.2021.9427352","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427352","url":null,"abstract":"Tiny Machine Learning (TinyML) is an emerging concept that concerns the execution of ML tasks on very constrained IoT devices. Although TinyML has generated a strong R&D interest around it, various challenges limit its effective execution in the constrained devices world, with the result of slowing down the development of a complete ecosystem around it. TinyML as-a-Service (TinyMLaaS) aims to fill the gap in this respect, with the definition of a set of guidelines that can enable an easier democratization of TinyML. In this paper, we describe how the \"as-a-Service\" model is bound to TinyML, by providing an overview of our concept and introducing the design requirements and building blocks that can make TinyMLaaS reality.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126321984","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Machine Learning System-Enabled GPU Acceleration for EDA","authors":"Tsung-Wei Huang","doi":"10.1109/VLSI-DAT52063.2021.9427323","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427323","url":null,"abstract":"Electronic design automation (EDA) contains many of the most challenging computational problems that optimize circuit layouts across billions of transistors. Most existing EDA algorithms are architecturally constrained by CPU parallelism, and their performances stagnate at about 8-16 cores To achieve new transformational performance milestones, new EDA algorithms must harness the power of heterogeneous parallelism comprising manycore CPUs and GPUs. However, these milestones are too difficult to achieve without a suitable software system to assist developers in the implementation complexities built into heterogeneous parallelism. Thanks to the advancement of machine learning, various software systems have been introduced to streamline the design of heterogeneous programs, and these results have inspired new heterogeneous parallel EDA solutions. In this talk, we present our research on GPU acceleration for placement and timing analysis by harnessing the power of machine learning systems. As an example, we achieve 500× speed-up for static timing analysis on a million-gate design.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131125216","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Reconfigurable Database Processor for Query Acceleration on FPGA","authors":"Bo-En Chen, Bo-Yen Lin, B. Lai","doi":"10.1109/VLSI-DAT52063.2021.9427326","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427326","url":null,"abstract":"FPGA (Field Programmable Gate Array) has been adopted as an efficient acceleration platform to support high throughput queries to big database. This paper proposes a Reconfigurable Database Processor (RDP) to enable end-to-end query processing. The engines can be reconfigured to support different design parameters. The data processing modules are allocated into stages, and composed to generate the accelerator structure for queries. We further proposed a systematic flow to adapt the design of RDP to the requirements of different queries. The experiments on TPC-DS have shown that RDP attains 56.5% higher throughput over the conventional solution on CPU. When compared with the previous FPGA designs, RDP shows 23.8% advantage on the overall performance while saving 13.6% FPGA resources.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133243624","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Po-Hsin Lin, Chang-Lin Shih, Davy P. Y. Wong, Pai H. Chou
{"title":"Gait Parameters Analysis Based on Leg-and-shoe-mounted IMU and Deep Learning","authors":"Po-Hsin Lin, Chang-Lin Shih, Davy P. Y. Wong, Pai H. Chou","doi":"10.1109/VLSI-DAT52063.2021.9427325","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427325","url":null,"abstract":"We propose a wearable system and a chain of methods for estimating stride lengths from inertial measurement units (IMU). The data are first processed by a Long Short-Term Memory (LSTM)-based method to determine the timing of step events. The raw IMU data and extracted features are also fed to LSTM to construct a regression model for learning stride lengths. Experimental results show that the proposed step event detector can reach −0.015 s to 0.02 s accuracy, and the proposed precision of stride-length estimator achieves 3.8 cm for both the mechanical model and LSTM model with extracted features.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124641266","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
K. Chen, Chun-Chuan Wang, Cheng-Kang Tsai, Jing-Wen Liang
{"title":"Dynamic Mapping Mechanism to Compute DNN Models on a Resource-limited NoC Platform","authors":"K. Chen, Chun-Chuan Wang, Cheng-Kang Tsai, Jing-Wen Liang","doi":"10.1109/VLSI-DAT52063.2021.9427320","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427320","url":null,"abstract":"The conventional Deep Neural Networks (DNNs) accelerator is usually designed based on array-based processing element computing. By using the array-based computation, memory access can be reduced efficiently based on a specific dataflow. However, the computing flexibility of the contemporary DNN accelerators is usually restricted because they support the monotonous computing dataflow. Hence, the computing efficiency will be degraded under the different hyperparameters of the target DNN (e.g., kernel size, layers, etc.) Because of the high flexibility and scalability of Network on Chip (NoC) interconnection, the NoC-based DNN design methodology becomes an attractive design paradigm, which reduces the design complexity of the DNN accelerator implementation. However, the current NoC-based DNN designs usually assume that the entire DNN model can be mapped to the target NoC. In this way, the area overhead will be larger with respect to the increasing DNN scale. To solve this problem, we propose a dynamic mapping algorithm, called: dense mapping. The dense mapping is used to map the neuron operations to the NoC as long as the available computing resources are enough for the neuron operations. Besides, an input sharing mechanism is proposed to reuse input data. In this way, we can not only process a DNN model on a small-scale NoC but also decrease the number of memory access by using the proposed input sharing mechanism. Compared with the related work, the proposed approaches help to reduce 60.5% number of memory access and improve 96.5% throughput due to fewer memory accesses.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"54 ","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131692827","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Opportunity and Challenge of Chiplet-Based HPC and AIoT","authors":"J. Lai","doi":"10.1109/VLSI-DAT52063.2021.9427309","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427309","url":null,"abstract":"High-performance computing and AI tremendously drive technology innovations on architecture, algorithm, memory, and semiconductor design, and continuously impact many fields from computation-intensive applications such as advanced driver-assistance system (ADAS) to highly-heterogeneous integrated, while performance-sensitive systems like 5G mobile and AIoT. Chiplet concept is considered a promising approach to alleviate the pressure on performance, power, cost, yield, and time to market raising with the increasing complexity of the fore-mentioned technologies. In this paper, we like to discuss the challenges and opportunities to satisfy these demands in conjunction with chiplet techniques.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115949307","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A Hybrid Supply Modulator for 10-MHz LTE Power Amplifier with 17.3% PAE Improvement","authors":"Y. Chen, M. Chiu, How-Wei Teng, Tsung-Hsien Lin","doi":"10.1109/VLSI-DAT52063.2021.9427340","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427340","url":null,"abstract":"A hybrid supply modulator (HSM) is proposed for the envelope-tracking power amplifier (ET-PA) to improve its power- added efficiency (PAE). The proposed HSM consists of a linear amplifier (LA) and a switching amplifier (SA). The LA adopts a current-recycling OTA (CROTA) followed by an IQ-controlled class- AB output stage. The CROTA enhances gain and slew rate of the LA first stage. In the class-AB output stage, the quiescent current (IQ) is regulated to reduce its sensitivity to process and supply variations. The SA employs a zero-peaked hysteresis comparator for switching control, which reduces the time delay of the SA control loop, and leads to better power efficiency. Measurement results shows that the HSM reaches 75.2% efficiency at 29.3-dBm output power. When tested with a commercial PA, the PAE improves by 17.3% at 22-dBm RF output power with 34.9-dB power gain.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115310341","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Holistic and In-Context Design Flow for 2.5D Chiplet-Package Interaction Co-Optimization","authors":"M. Kabir, Weishiun Hung, Tsung-Yi Ho, Yarui Peng","doi":"10.1109/VLSI-DAT52063.2021.9427353","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427353","url":null,"abstract":"In recent days, 2.5D package designs have gained popularity with an increasing number of heterogeneous chiplets integrated into advanced system-in-packages. RDL wires become longer and denser, presenting a growing impact on system performance, reliability, and integrity. At present, there exists no standard CAD flow that can design, analyze, and optimize a complete heterogeneous 2.5D system. The traditional die-by-die design approach processes each component independently during extraction and optimization and cannot be applied to heterogeneous systems without fundamental changes in standard CAD tools. Not only the chiplet-package extraction is inaccurate between the die-package interface ignoring all RDL capacitive and inductive impacts, but traditional CAD tools are also unable to perform cross-boundary design optimization.(p/)(p)We present a complete chiplet-package co-optimization flow for both homogeneous and heterogeneous 2.5D designs. It encompasses 2.5D-aware partitioning, chiplet-package co-planning, holistic and in-context extraction, package inductance consideration, and iterative optimization, along with design analysis and verification of the entire 2.5D system. In our previous work [1] targeting heterogeneous systems, we achieved an extraction error ranging between -2.10% and 24.0%. The in-context design flow proposed in this work achieves less than 1% extraction error on ground and coupling capacitance. This extraction result can be used to perform timing analysis with 99.8% accuracy and to generate timing context with 99.4% accuracy for iterative optimization.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117291914","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"An 8-Bit 1.25-GS/s 2.5-GHz ERBW Folding-Subrange ADC with Power-Efficient Metastability Error Reduction Technique","authors":"B. Chen, Yung-Hui Chung, Chia-Ming Tsai","doi":"10.1109/VLSI-DAT52063.2021.9427312","DOIUrl":"https://doi.org/10.1109/VLSI-DAT52063.2021.9427312","url":null,"abstract":"This paper presents an 8-bit 1.25GS/s folding-subrange ADC, implemented in 65nm CMOS technology. We design a coarse comparator to relieve critical metastability issue. A latch sharing technique in fine comparator can further reduce area overhead. Operating at 1.25GHz sampling rate with low input frequency, the measured SFDR and SNDR are 56.5dB and 43.4dB, respectively. When split-then-share buffer turns on, ERBW achieves up to 2.5 GHz within ±1dB bandwidth flatness. Peak DNL and INL are 0.71LSB and 0.72LSB, respectively. Setting at error magnitude of 4LSB and 32LSB, the bit error rate (BER) can achieve 5.4×10−6 and 5×10−10 errors/conversion, respectively. The ADC core consumes 20mW at 1.2V supply and occupies an active area of 0.14mm2.","PeriodicalId":362923,"journal":{"name":"2021 International Symposium on VLSI Design, Automation and Test (VLSI-DAT)","volume":"94 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124598936","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}