Pub Date : 2022-04-01DOI: 10.1109/CICC53496.2022.9772797
M. Beikmirza, Yiyu Shen, L. D. Vreede, M. Alavi
To support wideband complex modulated signals and comply with the stringent requirements of modern communication standards in an energy-efficient manner, recently, digital transmitters (DTXs) have been explored to fully benefit from the high-speed switching and integration capabilities of nanoscale CMOS technologies [1]–[5]. These DTXs are primarily exploiting a polar or Cartesian architecture. In a polar DTX [1], [2], two eigenvectors of amplitude (p) and phase $(phi)$ are generated from the in-phase (I) and quadrature (Q) baseband signals using non-linear coordinate rotation transformations (i.e., CORDIC). Provided that $rho$ is constant, the achievable drain efficiency (DE) is constant (Fig. 1 top). However, polar $text{DTXs}$ cannot manage large modulation bandwidth due to their non-linear $mathrm{I}/mathrm{Q}$ to $rho/phi$ conversion. Moreover, their phase and amplitude paths must recombine at the output stage without any delay mismatch to maintain linear operation. In contrast, Cartesian DTX variants can handle signals with large modulation bandwidth [3]. Nevertheless, their DE is lower than their polar counterparts owing to the linear combination of orthogonal I/Q vectors, yielding a 3-dB worst-case output power loss at the orthogonal (I/Q) axes. Alternatively, a multi-phase operation can be utilized that compromises polar and Cartesian features by mapping the I/Q signals into two non-orthogonal basis vectors with 45° relative phase difference and magnitudes of $mathrm{I}_{text{MP}}=mathrm{I}-mathrm{Q}$, QMP $=sqrt{2}mathrm{Q}$ [4]. This architecture inherits the advantages of the cartesian DTX, such as wideband operation, symmetrical, and synchronized $mathrm{I} /mathrm{Q}$ paths along with a DE behavior that imitates the polar case. This paper presents a multi-mode DTX that uses both Cartesian and multi-phase operation modes to target applications requiring large modulation bandwidth, decent spectral purity and average efficiency.
{"title":"A 1-to-4GHz Multi-Mode Digital Transmitter in 40nm CMOS Supporting 200MHz 1024-QAM OFDM signals with more than 23dBm/66% Peak Power/Drain Efficiency","authors":"M. Beikmirza, Yiyu Shen, L. D. Vreede, M. Alavi","doi":"10.1109/CICC53496.2022.9772797","DOIUrl":"https://doi.org/10.1109/CICC53496.2022.9772797","url":null,"abstract":"To support wideband complex modulated signals and comply with the stringent requirements of modern communication standards in an energy-efficient manner, recently, digital transmitters (DTXs) have been explored to fully benefit from the high-speed switching and integration capabilities of nanoscale CMOS technologies [1]–[5]. These DTXs are primarily exploiting a polar or Cartesian architecture. In a polar DTX [1], [2], two eigenvectors of amplitude (p) and phase $(phi)$ are generated from the in-phase (I) and quadrature (Q) baseband signals using non-linear coordinate rotation transformations (i.e., CORDIC). Provided that $rho$ is constant, the achievable drain efficiency (DE) is constant (Fig. 1 top). However, polar $text{DTXs}$ cannot manage large modulation bandwidth due to their non-linear $mathrm{I}/mathrm{Q}$ to $rho/phi$ conversion. Moreover, their phase and amplitude paths must recombine at the output stage without any delay mismatch to maintain linear operation. In contrast, Cartesian DTX variants can handle signals with large modulation bandwidth [3]. Nevertheless, their DE is lower than their polar counterparts owing to the linear combination of orthogonal I/Q vectors, yielding a 3-dB worst-case output power loss at the orthogonal (I/Q) axes. Alternatively, a multi-phase operation can be utilized that compromises polar and Cartesian features by mapping the I/Q signals into two non-orthogonal basis vectors with 45° relative phase difference and magnitudes of $mathrm{I}_{text{MP}}=mathrm{I}-mathrm{Q}$, QMP $=sqrt{2}mathrm{Q}$ [4]. This architecture inherits the advantages of the cartesian DTX, such as wideband operation, symmetrical, and synchronized $mathrm{I} /mathrm{Q}$ paths along with a DE behavior that imitates the polar case. This paper presents a multi-mode DTX that uses both Cartesian and multi-phase operation modes to target applications requiring large modulation bandwidth, decent spectral purity and average efficiency.","PeriodicalId":415990,"journal":{"name":"2022 IEEE Custom Integrated Circuits Conference (CICC)","volume":"112 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134577059","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}
Pub Date : 2022-04-01DOI: 10.1109/CICC53496.2022.9772869
Xin Ming, Zhikang Lin, Tian-yi Sun, Yao Qin, Yuan-Yuan Liu, Chun-wang Zhuang, Zhaoji Li, Bo Zhang
For future automotive applications, the growing demand for tiny, high power density and fast dynamic response is putting more pressure on power converters, where Gallium nitride FETs have proven to be promising devices [1]. However, for high conversion-ratio GaN power converters, the floating power rail control of half-bridge gate driver and low FOM/high-reliability level shifter (LS) pose a big challenge when increasing switching speed dV/dt and frequency (smaller Ton, min and Toff, min). Charging saturation and over-voltage protection of the bootstrap power supply, as well as common-mode transient immunity (CMTI)/transmission delay/power consumption of LS may introduce efficiency degradation and significant reliability issues.
{"title":"An Up to 10MHz 6.8% Minimum Duty Ratio GaN Driver with Dual-MOS-Switches Bootstrap and Adaptive Short-Pulse Based High-CMTI Level Shifter Achieving 6.05% Efficiency Improvement","authors":"Xin Ming, Zhikang Lin, Tian-yi Sun, Yao Qin, Yuan-Yuan Liu, Chun-wang Zhuang, Zhaoji Li, Bo Zhang","doi":"10.1109/CICC53496.2022.9772869","DOIUrl":"https://doi.org/10.1109/CICC53496.2022.9772869","url":null,"abstract":"For future automotive applications, the growing demand for tiny, high power density and fast dynamic response is putting more pressure on power converters, where Gallium nitride FETs have proven to be promising devices [1]. However, for high conversion-ratio GaN power converters, the floating power rail control of half-bridge gate driver and low FOM/high-reliability level shifter (LS) pose a big challenge when increasing switching speed dV/dt and frequency (smaller Ton, min and Toff, min). Charging saturation and over-voltage protection of the bootstrap power supply, as well as common-mode transient immunity (CMTI)/transmission delay/power consumption of LS may introduce efficiency degradation and significant reliability issues.","PeriodicalId":415990,"journal":{"name":"2022 IEEE Custom Integrated Circuits Conference (CICC)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133380916","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}
Pub Date : 2022-04-01DOI: 10.1109/CICC53496.2022.9772809
Shiyu Su, M. Chen
High-performance digital-to-analog converter (DAC) is demanded in many electronic systems requiring the synthesis of a high dynamic-range wideband signal. In the real implementation of a DAC, its sample rate and dynamic range (determined by linearity and noise) typically tradeoff with each other, limiting the achievable dynamic range for high-speed operation. This paper reviews recent advances in DAC architectures and discusses various relevant circuit and signal processing techniques that allow a DAC to potentially achieve a high speed and high dynamic range simultaneously. Lastly, findings relating to recent DAC silicon prototypes aiming in this direction will also be reviewed.
{"title":"High-Speed Digital-to-Analog Converter Design Towards High Dynamic Range","authors":"Shiyu Su, M. Chen","doi":"10.1109/CICC53496.2022.9772809","DOIUrl":"https://doi.org/10.1109/CICC53496.2022.9772809","url":null,"abstract":"High-performance digital-to-analog converter (DAC) is demanded in many electronic systems requiring the synthesis of a high dynamic-range wideband signal. In the real implementation of a DAC, its sample rate and dynamic range (determined by linearity and noise) typically tradeoff with each other, limiting the achievable dynamic range for high-speed operation. This paper reviews recent advances in DAC architectures and discusses various relevant circuit and signal processing techniques that allow a DAC to potentially achieve a high speed and high dynamic range simultaneously. Lastly, findings relating to recent DAC silicon prototypes aiming in this direction will also be reviewed.","PeriodicalId":415990,"journal":{"name":"2022 IEEE Custom Integrated Circuits Conference (CICC)","volume":"54 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114365083","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}
Pub Date : 2022-04-01DOI: 10.1109/CICC53496.2022.9772843
Mingqiang Guo, Sai-Weng Sin, Liang Qi, Gang Xiao, R. Martins
A SAR ADC comprises only a T/H, a comparator, SAR logics, and a capacitive DAC, thus exhibiting a power-efficient topology with low complexity, low power consumption, and friendly process technology scaling down. Consequently, it has a wide utilization in high-speed applications (like in time-interleaved SARs). Previous works improved the 1b/cycle topology to speed up SAR ADC conversions, leading to multi-bit/cycle [1] and N-bits N-comparators [2] structures. Compared with the above architectures, the conventional 1b/cycle topology still has apparent advantages related to low complexity, less parasitic, and less offset problems. Therefore, currently, the 1b/cycle is still the first choice for the majority of high-speed TI SAR ADCs [3]. The popularization of the high-speed SAR ADC with redundant bit structures can lead to a very short settling time required for the DACs [4]. However, the speed of the SAR is still a bottleneck, especially limited by the digital SAR logic [2].
SAR ADC仅由T/H、比较器、SAR逻辑和电容式DAC组成，因此具有低复杂度、低功耗和友好的工艺技术。因此，它在高速应用(如时间交错sar)中有广泛的应用。先前的工作改进了1b/周期拓扑以加快SAR ADC转换，从而产生了多位/周期[1]和n位n比较器[2]结构。与上述架构相比，传统的1b/cycle拓扑结构在低复杂度、少寄生、少偏移等方面仍然具有明显的优势。因此，目前，1b/周期仍然是大多数高速TI SAR adc的首选[3]。具有冗余位结构的高速SAR ADC的普及可以使dac的建立时间非常短[4]。然而，SAR的速度仍然是一个瓶颈，特别是受到数字SAR逻辑的限制[2]。
{"title":"A 10b 700MS/s single-channel 1b/cycle SAR ADC using a monotonic-specific feedback SAR logic with power-delay-optimized unbalanced N/P-MOS sizing","authors":"Mingqiang Guo, Sai-Weng Sin, Liang Qi, Gang Xiao, R. Martins","doi":"10.1109/CICC53496.2022.9772843","DOIUrl":"https://doi.org/10.1109/CICC53496.2022.9772843","url":null,"abstract":"A SAR ADC comprises only a T/H, a comparator, SAR logics, and a capacitive DAC, thus exhibiting a power-efficient topology with low complexity, low power consumption, and friendly process technology scaling down. Consequently, it has a wide utilization in high-speed applications (like in time-interleaved SARs). Previous works improved the 1b/cycle topology to speed up SAR ADC conversions, leading to multi-bit/cycle [1] and N-bits N-comparators [2] structures. Compared with the above architectures, the conventional 1b/cycle topology still has apparent advantages related to low complexity, less parasitic, and less offset problems. Therefore, currently, the 1b/cycle is still the first choice for the majority of high-speed TI SAR ADCs [3]. The popularization of the high-speed SAR ADC with redundant bit structures can lead to a very short settling time required for the DACs [4]. However, the speed of the SAR is still a bottleneck, especially limited by the digital SAR logic [2].","PeriodicalId":415990,"journal":{"name":"2022 IEEE Custom Integrated Circuits Conference (CICC)","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125206584","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}
High-speed time-interleaved ADCs are becoming more common in wireline receiver front-ends due to the enabling of subsequent digital processing for equalization and easier support of higher-order modulation schemes [1]. As technology nodes scale, ADCs based on the digital-intensive SAR architecture are more pervasive. However, implementations with the most common SAR algorithm that has sequential single-bit conversion cycles can result in large time-interleaving factors. Also, the sampling of wideband analog signals associated with higher data rates is difficult for conventional bootstrapped switch (BS) T/H circuits that have not adequately scaled in performance. One reason for this is that the low-duty-cycle sampling clocks, which are utilized for avoiding sampling crosstalk between time-interleaved sub-ADCs, shorten the tracking time and requires improvements in T/H circuit startup time. This motivates the use of simple NMOS switches in high-speed ADCs [2], [3]. However, this can negatively impact the high-speed linearity and ADC front-end bandwidth and also require higher supply voltages. This paper presents an ADC that utilizes both a high-bandwidth interleaver architecture based on a speed-enhanced bootstrapped switch and a pipelined-SAR unit ADC with output level shifting (OLS) settling [4] to enable low-power high-speed operation. At 38GS/s, the 7b ADC achieves 41.9fJ/conv.-step at low input frequencies, 64.1fJ/conv.-step at Nyquist, and has 20GHz 3dB bandwidth.
{"title":"A 38GS/s 7b Time-Interleaved Pipelined-SAR ADC with Speed-Enhanced Bootstrapped Switch in 22nm FinFET","authors":"Yuanming Zhu, Tong Liu, Srujan Kumar Kaile, Shiva Kiran, II-Min Yi, Ruida Liu, Julian Camilo Gomez Diaz, S. Hoyos, S. Palermo","doi":"10.1109/CICC53496.2022.9772785","DOIUrl":"https://doi.org/10.1109/CICC53496.2022.9772785","url":null,"abstract":"High-speed time-interleaved ADCs are becoming more common in wireline receiver front-ends due to the enabling of subsequent digital processing for equalization and easier support of higher-order modulation schemes [1]. As technology nodes scale, ADCs based on the digital-intensive SAR architecture are more pervasive. However, implementations with the most common SAR algorithm that has sequential single-bit conversion cycles can result in large time-interleaving factors. Also, the sampling of wideband analog signals associated with higher data rates is difficult for conventional bootstrapped switch (BS) T/H circuits that have not adequately scaled in performance. One reason for this is that the low-duty-cycle sampling clocks, which are utilized for avoiding sampling crosstalk between time-interleaved sub-ADCs, shorten the tracking time and requires improvements in T/H circuit startup time. This motivates the use of simple NMOS switches in high-speed ADCs [2], [3]. However, this can negatively impact the high-speed linearity and ADC front-end bandwidth and also require higher supply voltages. This paper presents an ADC that utilizes both a high-bandwidth interleaver architecture based on a speed-enhanced bootstrapped switch and a pipelined-SAR unit ADC with output level shifting (OLS) settling [4] to enable low-power high-speed operation. At 38GS/s, the 7b ADC achieves 41.9fJ/conv.-step at low input frequencies, 64.1fJ/conv.-step at Nyquist, and has 20GHz 3dB bandwidth.","PeriodicalId":415990,"journal":{"name":"2022 IEEE Custom Integrated Circuits Conference (CICC)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128629604","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}
Pub Date : 2022-04-01DOI: 10.1109/CICC53496.2022.9772827
Dhruv Rajendra Patel, A. Sharif-Bakhtiar, A. C. Carusone
Low-cost optical receivers (RX) operating at 100+ Gb/s 4-PAM with low power are in high demand to support 400GBASE-DR4/FR4 links in data centers. Existing pluggable solutions generally realize the RX front-end in BiCMOS. However, a more integrated solution, with the RX front-ends integrated onto a CMOS host IC and co-packaged alongside the photodiodes (PDs), offers the potential for smaller size, lower cost, and lower power [1], [2]. This work demonstrates a 112 Gb/s 4-PAM linear TIA in CMOS flip-chip co-packaged with commercial PDs and different PD-to-RX interconnect lengths (Fig. 1a).
{"title":"A 112 Gb/s -8.2 dBm Sensitivity 4-PAM Linear TIA in 16nm CMOS with Co-Packaged Photodiodes","authors":"Dhruv Rajendra Patel, A. Sharif-Bakhtiar, A. C. Carusone","doi":"10.1109/CICC53496.2022.9772827","DOIUrl":"https://doi.org/10.1109/CICC53496.2022.9772827","url":null,"abstract":"Low-cost optical receivers (RX) operating at 100+ Gb/s 4-PAM with low power are in high demand to support 400GBASE-DR4/FR4 links in data centers. Existing pluggable solutions generally realize the RX front-end in BiCMOS. However, a more integrated solution, with the RX front-ends integrated onto a CMOS host IC and co-packaged alongside the photodiodes (PDs), offers the potential for smaller size, lower cost, and lower power [1], [2]. This work demonstrates a 112 Gb/s 4-PAM linear TIA in CMOS flip-chip co-packaged with commercial PDs and different PD-to-RX interconnect lengths (Fig. 1a).","PeriodicalId":415990,"journal":{"name":"2022 IEEE Custom Integrated Circuits Conference (CICC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131099833","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}
Pub Date : 2022-04-01DOI: 10.1109/CICC53496.2022.9772777
Bingzheng Yang, H. Qian, Yiyang Shu, Jie Zhou, Xun Luo
To meet the demands of multi-Gb/s data-rate wireless transmission such as millimeter wave (mmW) 5G wireless, there is a growing requirement of mmW transmitters (TXs) with high output power, high efficiency, and high data-rate [1], [2]. The digital power amplifier (DPA) shows the merits of high system efficiency, high integration with baseband, and low cost compared with analog PAs [3], which is potential for mmW 5G TX. However, the increasing parasitic of transistor at mmW limits switching speed, which leads to efficiency degradation [4]–[6]. Meanwhile, the routing parasitics deteriorate the output power and efficiency of mmW DPA [5]. DPA usually suffers from relatively large LO leakage at mmW, which decreases the dynamic range and introduces linearity issue. To overcome these challenges, a 2×9-bit quadrature hybrid switched-capacitor PA (SCPA) with LO leakage self-suppression and distributed parasitic-cancelling sub-PA array is proposed, which achieves saturated Pout of 24.03dBm with peak system efficiency (SE) of 31.5% at 24GHz.
{"title":"22-30GHz Quadrature Hybrid SCPA with LO Leakage Self-Suppression and Distributed Parasitic-Cancelling Sub-PA Array for Linearity and Efficiency Enhancement","authors":"Bingzheng Yang, H. Qian, Yiyang Shu, Jie Zhou, Xun Luo","doi":"10.1109/CICC53496.2022.9772777","DOIUrl":"https://doi.org/10.1109/CICC53496.2022.9772777","url":null,"abstract":"To meet the demands of multi-Gb/s data-rate wireless transmission such as millimeter wave (mmW) 5G wireless, there is a growing requirement of mmW transmitters (TXs) with high output power, high efficiency, and high data-rate [1], [2]. The digital power amplifier (DPA) shows the merits of high system efficiency, high integration with baseband, and low cost compared with analog PAs [3], which is potential for mmW 5G TX. However, the increasing parasitic of transistor at mmW limits switching speed, which leads to efficiency degradation [4]–[6]. Meanwhile, the routing parasitics deteriorate the output power and efficiency of mmW DPA [5]. DPA usually suffers from relatively large LO leakage at mmW, which decreases the dynamic range and introduces linearity issue. To overcome these challenges, a 2×9-bit quadrature hybrid switched-capacitor PA (SCPA) with LO leakage self-suppression and distributed parasitic-cancelling sub-PA array is proposed, which achieves saturated Pout of 24.03dBm with peak system efficiency (SE) of 31.5% at 24GHz.","PeriodicalId":415990,"journal":{"name":"2022 IEEE Custom Integrated Circuits Conference (CICC)","volume":"134 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127102962","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}
Pub Date : 2022-04-01DOI: 10.1109/CICC53496.2022.9772844
M. Rodwell, Ahmed S. H. Ahmed, M. Seo, Utku Soylu, A. Alizadeh, Navid Hosseinzadeh
100–300 GHz wireless systems can provide very high data rates per signal beam, and, given the short wavelengths, even compact arrays can contain many elements, and hence can simultaneously transmit, in the same frequency band, many simultaneous independent signal beams to further greatly increase capacity. We will describe representative system designs, including wireless hubs and backhaul links using massive spatial multiplexing, plus imaging radar systems, evaluate their feasible performance, and identify the key challenges in implementation, including transistor and IC performance, array physical design, digital beam former complexity, and systems cost.
{"title":"IC and Array Technologies for 100-300GHz Wireless","authors":"M. Rodwell, Ahmed S. H. Ahmed, M. Seo, Utku Soylu, A. Alizadeh, Navid Hosseinzadeh","doi":"10.1109/CICC53496.2022.9772844","DOIUrl":"https://doi.org/10.1109/CICC53496.2022.9772844","url":null,"abstract":"100–300 GHz wireless systems can provide very high data rates per signal beam, and, given the short wavelengths, even compact arrays can contain many elements, and hence can simultaneously transmit, in the same frequency band, many simultaneous independent signal beams to further greatly increase capacity. We will describe representative system designs, including wireless hubs and backhaul links using massive spatial multiplexing, plus imaging radar systems, evaluate their feasible performance, and identify the key challenges in implementation, including transistor and IC performance, array physical design, digital beam former complexity, and systems cost.","PeriodicalId":415990,"journal":{"name":"2022 IEEE Custom Integrated Circuits Conference (CICC)","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121459644","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}
Pub Date : 2022-04-01DOI: 10.1109/CICC53496.2022.9772817
Naresh R Shanbhag, Saion K. Roy
Since its inception in 2014 [1], the modern version of in-memory computing (IMC) has become an active area of research in integrated circuit design globally for realizing artificial intelligence and machine learning workloads. Since 2018, > 40 IMC-related papers have been published in top circuit design conferences demonstrating significant reductions (>20X) in energy over their digital counterparts especially at the bank-level. Today, bank-level IMC designs have matured but it is not clear what the limiting factors are. This lack of clarity is due to multiple reasons including: 1) the conceptual complexity of IMCs due to its full-stack (devices-to-systems) nature, 2) the presence of a fundamental energy-efficiency vs. compute SNR trade-off due to its analog computations, and 3) the statistical nature of machine learning workloads. The absence of a rigorous benchmarking methodology for IMCs - a problem facing machine learning ICs in general [2] - further obfuscates the underlying trade-offs. As a result, it has become difficult to evaluate the novelty of IMC-related ideas being proposed and therefore gauge the true progress in this exciting field.
{"title":"Comprehending In-memory Computing Trends via Proper Benchmarking","authors":"Naresh R Shanbhag, Saion K. Roy","doi":"10.1109/CICC53496.2022.9772817","DOIUrl":"https://doi.org/10.1109/CICC53496.2022.9772817","url":null,"abstract":"Since its inception in 2014 [1], the modern version of in-memory computing (IMC) has become an active area of research in integrated circuit design globally for realizing artificial intelligence and machine learning workloads. Since 2018, > 40 IMC-related papers have been published in top circuit design conferences demonstrating significant reductions (>20X) in energy over their digital counterparts especially at the bank-level. Today, bank-level IMC designs have matured but it is not clear what the limiting factors are. This lack of clarity is due to multiple reasons including: 1) the conceptual complexity of IMCs due to its full-stack (devices-to-systems) nature, 2) the presence of a fundamental energy-efficiency vs. compute SNR trade-off due to its analog computations, and 3) the statistical nature of machine learning workloads. The absence of a rigorous benchmarking methodology for IMCs - a problem facing machine learning ICs in general [2] - further obfuscates the underlying trade-offs. As a result, it has become difficult to evaluate the novelty of IMC-related ideas being proposed and therefore gauge the true progress in this exciting field.","PeriodicalId":415990,"journal":{"name":"2022 IEEE Custom Integrated Circuits Conference (CICC)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132421412","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：