[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-125497-en":3,"doc-seo-125497-105":30,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},125497,13056703020460,"Valentina","https://ap-avatar.wpscdn.com/avatar/be000253dac470eee5d?_k=1778207105932848923",6,"Technology","Circuits-Informed Machine Learning Technique for Blind Open-Loop Digital Calibration of SAR ADC","SAR ADCs offer high energy efficiency at medium resolution, but at above 10-bit resolution and sampling rates over 100 MHz performance is constrained by sampling noise, comparator noise and offset, and reference ripple. Existing corrections often rely on prior error knowledge or analog-oriented solutions, limiting generality. This work introduces a circuits-informed supervised machine learning method for blind open-loop digital calibration that simultaneously suppresses multiple static and dynamic errors. The approach trains using aligned outputs from a low-speed reference SAR ADC copy and employs a shallow ANN with circuit-derived features to ensure robust convergence.","Circuits-Informed Machine Learning Technique for Blind Open-Loop Digital Calibration of SAR ADC  \nSumukh Bhanushali†, Debnath Maiti†, Phaneendra Bikkina‡, Esko Mikkola‡, Arindam Sanyal†  \n† Arizona State University,‡ Alphacore, Inc.  \nSAR architecture is widely adopted due to its high energy efficiency at medium resolution. At >10-bit resolution and sampling frequency >100MHz, performance of SAR ADC is limited by sampling noise, comparator noise and offset, and reference ripple. Several techniques exist to suppress these non-idealities, such as stochastic estimation for reducing comparator noise [1], kT/C noise cancellation [2], comparator offset calibration in loop-unrolled SAR [3], and reference ripple suppression [4-7] . These techniques are algorithm-based which target specific errors and require knowledge of the errors to correct them. In addition, some of these techniques need analog solutions such as low-noise amplifiers for kT/C noise cancellation and low-output resistance buffers for reference regulation. In contrast, this work proposes a machine-learning (ML) approach for blind digital calibration that simultaneously suppresses multiple errors without requiring prior knowledge of these errors. The key advantage of the proposed technique is that the ML model can correct even errors unforeseen during design and are generalizable to multiple architectures unlike algorithm-based techniques. In the proposed technique, a supervised ML model uses the difference between the main SAR ADC and a lowspeed reference SAR ADC to learn the errors in the main ADC and suppress them. A few recent works have applied ML to improve ADC performance [8-10] . Ref. [8] applies ML to correct mismatches in time-interleaved ADC, but only achieves a modest improvement in ENOB of ~1 .5-bits even when using a complex convolutional neural network with >13,000 parameters. Ref. [9] employs unsupervised ML on a split-ADC architecture to correct errors by forcing the two ADCs to match each other. However, the unsupervised approach may not always converge to the desired operating point and requires each ADC to follow a different conversion trajectory that is neither trivial to realize nor easily generalizable to different architectures. Ref. [10] demonstrates supervised ML calibration on a 1MHz SAR ADC that does not capture errors introduced at high-speed, e.g. reference ripple. In addition, the reference ADC model is extracted in software from the main ADC which limits applicability of the calibration technique. In contrast, the proposed ML technique requires 9x less parameters than the ML model in [8], will always converge due to supervised ML training, implements both reference and main ADCsin silicon and calibrates both static and dynamic errors introduced at high-speed. The proposed architecture is shown in Fig.1. The main ADC is a 12-bit, top-plate sampled synchronous ADC with 1-bit redundancy and unit capacitor of 0.6fF. The ADC uses bi-directional single-sided switching (BDSS) technique to reduce switching energy [10] . In this work, the reference ADC used for supervised ML training is a copy of the main ADC but runs at Fs/8 where the Fs is the sampling speed of the main ADC. The two ADCs do not need to be matched accurately since the ML model calibrates mismatches between the two ADCs as well. The sampling instants of the main and reference ADCs align every 8 samples of the main ADC, and the ML model is trained during these aligned sampling instants. To reduce the training requirement, the ML model is trained to estimate only the errors between the main and reference ADC outputs rather than letting the ML model learn the main ADC transfer function. While the training happens only when the sampling instantsof the main and reference ADCs are aligned, the trained ML model performs calibration for all the main ADC outputs. In-order to achieve high SFDR without requiring a complex deep neural network for calibration, this work proposes a circuits-inf","cbCaik6Ba3itg1lU","https://ap.wps.com/l/cbCaik6Ba3itg1lU","pdf",1299209,1,3,"English","en",105,"# Blind open-loop digital calibration of SAR ADC\n## Problem: non-idealities at high resolution and high sampling rate\n## Proposed ML calibration concept and advantages\n## Training data and reference ADC architecture\n## Circuits-informed ANN feature design\n## Calibration of static and dynamic errors","[{\"question\":\"Why do SAR ADC performance degrade at high resolution and high sampling frequency?\",\"answer\":\"At \\u003e10-bit resolution and \\u003e100 MHz, sampling noise, comparator noise/offset, and reference ripple become dominant, limiting achievable performance.\"},{\"question\":\"How does the proposed method enable blind calibration without prior error knowledge?\",\"answer\":\"A supervised ML model learns errors by comparing outputs from the main SAR ADC and a low-speed reference SAR ADC copy, allowing correction of multiple errors even if some were unforeseen during design.\"},{\"question\":\"What architectural choices are used for training and calibration?\",\"answer\":\"The reference ADC is a copy of the main ADC running at Fs/8; the two align every 8 samples for training. A shallow ANN is trained to estimate inter-ADC errors using circuit-informed static and dynamic features, then applied to calibrate all main ADC outputs.\"}]","Circuits-Informed Machine Learning Technique for Blind Open-Loop Digital Calibration of SAR ADC | PDF",1785899340,8,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":28},"circuits-informed-machine-learning-technique-for-blind-open-loop-digital-calibration-of-sar-adc","",{"@graph":36,"@context":84},[37,53,67],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":21},"https://docshare.wps.com/document/technology/",{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/circuits-informed-machine-learning-technique-for-blind-open-loop-digital-calibration-of-sar-adc/125497/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-08-05",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"Why do SAR ADC performance degrade at high resolution and high sampling frequency?","Question",{"text":74,"@type":75},"At >10-bit resolution and >100 MHz, sampling noise, comparator noise/offset, and reference ripple become dominant, limiting achievable performance.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does the proposed method enable blind calibration without prior error knowledge?",{"text":79,"@type":75},"A supervised ML model learns errors by comparing outputs from the main SAR ADC and a low-speed reference SAR ADC copy, allowing correction of multiple errors even if some were unforeseen during design.",{"name":81,"@type":72,"acceptedAnswer":82},"What architectural choices are used for training and calibration?",{"text":83,"@type":75},"The reference ADC is a copy of the main ADC running at Fs/8; the two align every 8 samples for training. A shallow ANN is trained to estimate inter-ADC errors using circuit-informed static and dynamic features, then applied to calibrate all main ADC outputs.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,112,117,121,126,129,133],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":46,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":110,"slug":111},50,"technology",{"id":113,"doc_module":4,"doc_module_name":46,"category_name":114,"show_sort_weight":115,"slug":116},7,"Healthcare",40,"healthcare",{"id":29,"doc_module":4,"doc_module_name":46,"category_name":118,"show_sort_weight":119,"slug":120},"Research & Report",30,"research-report",{"id":122,"doc_module":4,"doc_module_name":46,"category_name":123,"show_sort_weight":124,"slug":125},9,"Religion & Spirituality",20,"religion-spirituality",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":127,"show_sort_weight":124,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":46,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":46,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]