Transformers

An Inductive-Load-Modulated Multi-Band Phase Shifter with <0.38°/0.12dB RMS Errors

An Inductive-Load-Modulated Multi-Band Phase Shifter with <0.38°/0.12dB RMS Errors 150 150

Abstract:

This paper presents a compact, multi-band Reflection-Type Phase Shifter (RTPS) implemented in 65-nm CMOS that overcomes the narrowband limitations of conventional passive loads. The proposed design utilizes an inductive load modulation. By injecting a secondary signal to actively manipulate the magnetic flux, the equivalent inductance is boosted to enable operation …

View on IEEE Xplore

ASAP: A 28-nm Transformer Training Accelerator With Alternating Sparsity and Asymmetrical Microscaling Precision

ASAP: A 28-nm Transformer Training Accelerator With Alternating Sparsity and Asymmetrical Microscaling Precision 150 150

Abstract:

This work presents ASAP, a 28-nm transformer-training accelerator that combines N:M structured sparsity with asymmetric microscaling floating-point (MXFP) precision through a unified algorithm–hardware co-design. ASAP introduces a progressive sparsity schedule in which pruned compute resources are reassigned to increase numerical precision for important weights and activations, stabilizing optimization …

View on IEEE Xplore

A BEV Perception Transformer Accelerator With Saliency-Driven Image/Point Cloud Fusion and Phase-Linked Dataflow in 28 nm CMOS

A BEV Perception Transformer Accelerator With Saliency-Driven Image/Point Cloud Fusion and Phase-Linked Dataflow in 28 nm CMOS 150 150

Abstract:

Deploying advanced Transformer-based models for real-time, high-accuracy multimodal bird’s-eye-view (BEV) perception in autonomous driving imposes substantial hardware demands. To address this, we propose a low-cost, low-power image/point-cloud fusion Transformer accelerator that supports two modes: high-performance driving and ultra-low-power sentry operation. We first propose a cross-modal saliency evaluation mechanism …

View on IEEE Xplore

MITTA: A Multi-Task Transformer Accelerator With Mixed Precision Structured Sparsity and Hierarchical Task-Adaptive Power Management

MITTA: A Multi-Task Transformer Accelerator With Mixed Precision Structured Sparsity and Hierarchical Task-Adaptive Power Management 150 150

Abstract:

This article presents MITTA, the first silicon-proven transformer accelerator optimized for multi-task inference across both natural language processing (NLP) and image processing domains. MITTA accelerates a task-sharing algorithm that minimizes sub-task computation by reusing both activations and weights from a shared base task, requiring only sparse delta computation for sub-tasks. …

View on IEEE Xplore

Design and Analysis of a 13.7–41 GHz Ultra-Wideband Frequency Doubler With Cross-Coupled Push-Push Structure

Design and Analysis of a 13.7–41 GHz Ultra-Wideband Frequency Doubler With Cross-Coupled Push-Push Structure 150 150

Abstract:

This article presents a 13.7–41 GHz ultra-wideband frequency doubler with high efficiency and conversion gain (CG). The proposed cross-coupled push-push structure, in conjunction with the fourth-order transformer-based resonator and the series gate inductor, collaboratively shapes the input signal amplitude such that three distinct peaks emerge at different frequencies, thereby significantly improving …

View on IEEE Xplore

A K/Ka-Band Transmit/Receive Front-End With Triple-Coupled Transformer Technique in 65-nm Bulk CMOS

A K/Ka-Band Transmit/Receive Front-End With Triple-Coupled Transformer Technique in 65-nm Bulk CMOS 150 150

Abstract:

This article presents a K/Ka-band transmit/receive (T/R) front-end for jointed sensing and communication (JSAC) applications. A reconfigurable matching network for both signal reception and transmission is realized using the proposed triple-coupled transformer (TCT) technique, achieving low power loss and a compact footprint. The T/R switch at …

View on IEEE Xplore

A Compact Reconfigurable Dual-Path Dual-Band LNA for 5G NR FR2 Applications

A Compact Reconfigurable Dual-Path Dual-Band LNA for 5G NR FR2 Applications 150 150

Abstract:

This article presents a reconfigurable dual-path dual-band low noise amplifier (LNA) for fifth generation (5G) millimeter-wave (mmW) communications. A novel band-switching input matching architecture based on the cross-connected transistors is proposed to achieve optimal dual-band input matching and $g_{m}$ -boosting. This architecture allows the dual-band input transistors to share …

View on IEEE Xplore

Analysis and Design of Power Amplifier Using Parallel-Combined Multisegment Transformer

Analysis and Design of Power Amplifier Using Parallel-Combined Multisegment Transformer 150 150

Abstract:

This letter presents a highly efficient power amplifier (PA) using a parallel-combined vertical multisegment transformer for 5G new radio (NR) applications operating in bands n257 and n258, in a 65-nm bulk CMOS process. A multisegment transformer facilitates a lower provided input impedance than a conventional transformer, enabling the PA to …

View on IEEE Xplore

A 3-D HBI Compliant 1.536 TB/s/mm2 Bandwidth Scalable Attention Accelerator With 22.5-GOPS Throughput High Speed SoftMax for Quantized Transformers in Intel 3

A 3-D HBI Compliant 1.536 TB/s/mm2 Bandwidth Scalable Attention Accelerator With 22.5-GOPS Throughput High Speed SoftMax for Quantized Transformers in Intel 3 150 150

Abstract:

This letter presents a novel hardware accelerator compatible with <3- $\mu $ m pitch 3-D Cu-Cu hybrid bonding interconnect (HBI) technology, particularly designed to efficiently execute multihead attention (MHA) of encoder transformer models. We present an accelerator that addresses performance losses due to low precision models by incorporating specialized hardware optimizations …

View on IEEE Xplore