Piergiulio Mannocci
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Piergiulio Mannocci has a total of 100 co-authors. Here are the top co-authors by number of shared publications:
- Daniele Ielmini
Dipartimento di Elettronica e Informazione and IU.NET, Politecnico di Milano, Milano, Italy
21 shared publications - Giacomo Pedretti
Hewlett Packard Labs, Hewlett Packard Enterprise, Milpitas, USA
9 shared publications - Zhong Sun
Peking University
8 shared publications - Matteo Farronato
Informazione e Bioingegneria (DEIB) Politecnico di Milano and IU.NET
5 shared publications - Artem Glukhov
Dipartimento di Elettronica, Informazione e Bioingegneria (DEIB), IU.NET, Politecnico di Milano, Milan, Italy
4 shared publications - Nicola Lepri
Dipartimento di Elettronica, Informazione e Bioingegneria (DEIB), Politecnico di Milano and IU.NET
4 shared publications
Latest Publications
- 14 January 2026
- Published by Springer Nature in Nature Electronics
- Vol. 9 (2) , 200-212
A fully integrated analogue closed-loop in-memory computing accelerator based on static random-access memory
- 14 January 2026
- Published by Springer Nature in Nature Electronics
- Vol. 9 (2) , 200-212
A fully integrated analogue closed-loop in-memory computing accelerator based on static random-access memory
- Top Cited PapersOpen Access
- 7 January 2026
- Published by Springer Nature in npj Unconventional Computing
- Vol. 3 (1) , 1
Achieving high precision in analog in-memory computing systems
Modern workloads challenge von Neumann architectures due to memory-processor data transfer. In-memory computing (IMC) enables in situ processing, with analog IMC (AIMC) based on resistive memories offering high-throughput and energy-efficient multiply-accumulate operations. Precision is limited by noise, device/circuit variations, and non-idealities. This work reviews error sources in AIMC and surveys mitigation strategies: bit slicing, residue number system, error correction codes, and mixed-precision iterative refinement, analyzing hardware implementations, overheads, and tradeoffs.
- 7 January 2026
- Published by Springer Nature in npj Unconventional Computing
- Vol. 3 (1) , 1
Top Cited PapersOpen AccessAchieving high precision in analog in-memory computing systems
Modern workloads challenge von Neumann architectures due to memory-processor data transfer. In-memory computing (IMC) enables in situ processing, with analog IMC (AIMC) based on resistive memories offering high-throughput and energy-efficient multiply-accumulate operations. Precision is limited by noise, device/circuit variations, and non-idealities. This work reviews error sources in AIMC and surveys mitigation strategies: bit slicing, residue number system, error correction codes, and mixed-precision iterative refinement, analyzing hardware implementations, overheads, and tradeoffs.
- Open Access
- 10 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 73 (1) , 325-331
High-Accuracy, High-Performance In-Memory Computing With High-Resistance Spin-Orbit Torque (SOT) Magnetic Memory
In-memory computing (IMC) has emerged as a promising solution for artificial intelligence (AI) accelerators due to the reduced data movement and improved parallelism in the crosspoint array. However, IMC faces several limitations such as the device variations affecting the computing accuracy, the area- and energy-consuming peripheral circuitry, and the time-consuming high-voltage program-verify operations of the nonvolatile memory (NVM). In addition, the relatively large summation currents cause IR drop within the array, leading to further loss of accuracy. To overcome these issues, this work presents 1-bit quantized networks based on spin-orbit-torque magnetoresistive random access-memory (SOT-MRAM) with high resistance–area (RA) product. We provide a detailed statistical characterization of SOT-MRAM arrays and develop quantization-aware training of various neural networks. Our results indicate that SOT-MRAM enables: 1) high inference accuracy, thanks to excellent uniformity; 2) negligible input-dependent IR drop nonlinearity, thanks to high resistance; and 3) high-speed, low-power reconfiguration, thanks to fast device programming. These results support high-RA SOT-MRAM for digital-like and reconfigurable IMC accelerators of edge AI.
- 10 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 73 (1) , 325-331
Open AccessHigh-Accuracy, High-Performance In-Memory Computing With High-Resistance Spin-Orbit Torque (SOT) Magnetic Memory
In-memory computing (IMC) has emerged as a promising solution for artificial intelligence (AI) accelerators due to the reduced data movement and improved parallelism in the crosspoint array. However, IMC faces several limitations such as the device variations affecting the computing accuracy, the area- and energy-consuming peripheral circuitry, and the time-consuming high-voltage program-verify operations of the nonvolatile memory (NVM). In addition, the relatively large summation currents cause IR drop within the array, leading to further loss of accuracy. To overcome these issues, this work presents 1-bit quantized networks based on spin-orbit-torque magnetoresistive random access-memory (SOT-MRAM) with high resistance–area (RA) product. We provide a detailed statistical characterization of SOT-MRAM arrays and develop quantization-aware training of various neural networks. Our results indicate that SOT-MRAM enables: 1) high inference accuracy, thanks to excellent uniformity; 2) negligible input-dependent IR drop nonlinearity, thanks to high resistance; and 3) high-speed, low-power reconfiguration, thanks to fast device programming. These results support high-RA SOT-MRAM for digital-like and reconfigurable IMC accelerators of edge AI.
- Open Access
- 4 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 72 (5) , 2677-2684
3-D Vertical Resistive Switching Random Access Memory (3D-VRRAM) With Multilevel Programming for High-Density, Energy-Efficient In-Memory Computing
Resistive random access memory (RRAM) devices offer a broad range of attractive properties for in-memory computing (IMC) applications, such as nonvolatile storage, low read current, and high scalability. IMC allows to overcome the memory bottleneck of data-intensive workloads, such as deep learning on the edge. In this context, 3-D vertical RRAM (3D-VRRAM) is a promising option to achieve high memory cell capacity with low fabrication cost. In this work, we present an HfOx-based 3D-VRRAM crossbar array (CBA) capable of IMC with precise multilevel programming. We show an extensive experimental demonstration of both matrix–vector multiplication (MVM) and inverse/pseudoinverse matrix calculation via IMC on 3D-VRRAM. To further support the parallel IMC application in real-life scenarios, the work also reports a demonstration of relatively large-size problems adopting 2D-RRAM and SRAM-based memory arrays. These results support 3D-VRRAM for high-density, energy-efficient IMC for edge computing applications.
- 4 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 72 (5) , 2677-2684
Open Access3-D Vertical Resistive Switching Random Access Memory (3D-VRRAM) With Multilevel Programming for High-Density, Energy-Efficient In-Memory Computing
Resistive random access memory (RRAM) devices offer a broad range of attractive properties for in-memory computing (IMC) applications, such as nonvolatile storage, low read current, and high scalability. IMC allows to overcome the memory bottleneck of data-intensive workloads, such as deep learning on the edge. In this context, 3-D vertical RRAM (3D-VRRAM) is a promising option to achieve high memory cell capacity with low fabrication cost. In this work, we present an HfOx-based 3D-VRRAM crossbar array (CBA) capable of IMC with precise multilevel programming. We show an extensive experimental demonstration of both matrix–vector multiplication (MVM) and inverse/pseudoinverse matrix calculation via IMC on 3D-VRRAM. To further support the parallel IMC application in real-life scenarios, the work also reports a demonstration of relatively large-size problems adopting 2D-RRAM and SRAM-based memory arrays. These results support 3D-VRRAM for high-density, energy-efficient IMC for edge computing applications.
- Open Access
- 29 March 2025
- Published by Springer Nature
- p. 45-54
Analog Circuit Design for In-Memory Linear Algebra Accelerators
Since its introduction in 1945, computing systems have been built around von Neumann’s architecture, predicating the physical separation of memory and computing units on grounds of flexibility and generality. However, the increasingly data-driven workloads of modern-day applications exacerbate the energy and latency overheads associated with data shuttling. In-memory computing (IMC) radically subverts the classical paradigm by performing computation in situ within the memory elements, unlocking theoretically unrivaled throughput and energy efficiency. Among the wide spectrum of IMC architectures, closed-loop in-memory computing (CL-IMC) has attracted interest for its capability to accelerate computationally heavy operations of increasing use, such as matrix inversion. This chapter focuses on analog closed-loop circuits for in-memory accelerators. A mathematical framework is derived to develop a matrix-based circuit simulator providing orders-of-magnitude speedups with respect to SPICE solvers. New circuits for the acceleration of regularized regressions and linear quadratic estimation are characterized in terms of accuracy and speed, providing improvement with respect to digital solvers in baseband processing in 6G systems and Kalman filters. Experimental demonstrations finally provide a real-world implementation of CL-IMC topologies. The obtained results strengthen the position of CL-IMC as a promising candidate for next-generation energy-efficient algebraic accelerators.
- 29 March 2025
- Published by Springer Nature
- p. 45-54
Open AccessAnalog Circuit Design for In-Memory Linear Algebra Accelerators
Since its introduction in 1945, computing systems have been built around von Neumann’s architecture, predicating the physical separation of memory and computing units on grounds of flexibility and generality. However, the increasingly data-driven workloads of modern-day applications exacerbate the energy and latency overheads associated with data shuttling. In-memory computing (IMC) radically subverts the classical paradigm by performing computation in situ within the memory elements, unlocking theoretically unrivaled throughput and energy efficiency. Among the wide spectrum of IMC architectures, closed-loop in-memory computing (CL-IMC) has attracted interest for its capability to accelerate computationally heavy operations of increasing use, such as matrix inversion. This chapter focuses on analog closed-loop circuits for in-memory accelerators. A mathematical framework is derived to develop a matrix-based circuit simulator providing orders-of-magnitude speedups with respect to SPICE solvers. New circuits for the acceleration of regularized regressions and linear quadratic estimation are characterized in terms of accuracy and speed, providing improvement with respect to digital solvers in baseband processing in 6G systems and Kalman filters. Experimental demonstrations finally provide a real-world implementation of CL-IMC topologies. The obtained results strengthen the position of CL-IMC as a promising candidate for next-generation energy-efficient algebraic accelerators.
- Open Access
- 17 January 2025
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 11 (3) , eadr3241
Seizure detection via reservoir computing in MoS 2 -based charge trap memory devices
Neurological disorders are a substantial global health burden, affecting millions of people worldwide. A key challenge in developing effective treatments and preventive measures is the realization of low-power wearable systems with early detection capabilities. Traditional strategies rely on machine learning algorithms, but their computational demands often exceed what miniaturized systems can provide. Neuromorphic computing, inspired by the human brain, demonstrated capabilities of on-chip computing with low power consumption. In this context, bidimensional (2D) semiconductors hold notable promise, thanks to their unique electronic properties, atomic-scale thickness, and scalability, making them ideal for low-power applications. This work presents a neuromorphic reservoir computing system exploiting MoS 2 -based charge trap memories (CTMs) for processing of electrophysiological signals. Real-time seizures detection is achieved, thanks to the nonlinear integration of local-field potential (LFP) recorded from in vitro rodent models of ictogenesis. The results support MoS 2 -based CTMs for low-power biomedical devices in clinical diagnosis and treatment of epilepsy.
- 17 January 2025
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 11 (3) , eadr3241
Open AccessSeizure detection via reservoir computing in MoS 2 -based charge trap memory devices
Neurological disorders are a substantial global health burden, affecting millions of people worldwide. A key challenge in developing effective treatments and preventive measures is the realization of low-power wearable systems with early detection capabilities. Traditional strategies rely on machine learning algorithms, but their computational demands often exceed what miniaturized systems can provide. Neuromorphic computing, inspired by the human brain, demonstrated capabilities of on-chip computing with low power consumption. In this context, bidimensional (2D) semiconductors hold notable promise, thanks to their unique electronic properties, atomic-scale thickness, and scalability, making them ideal for low-power applications. This work presents a neuromorphic reservoir computing system exploiting MoS 2 -based charge trap memories (CTMs) for processing of electrophysiological signals. Real-time seizures detection is achieved, thanks to the nonlinear integration of local-field potential (LFP) recorded from in vitro rodent models of ictogenesis. The results support MoS 2 -based CTMs for low-power biomedical devices in clinical diagnosis and treatment of epilepsy.
- Open Access
- 16 December 2024
- Published by Springer Nature in Research Square
An SRAM-based fully-integrated analog closed-loop in-memory computing accelerator
Conventional processors are unsuited for the increasingly data-intensive workloads brought upon by the resurgence of artificial intelligence (AI) and machine learning (ML) in the latest years. The widespread diffusion of ML techniques in many different technological fields has prompted both academia and industry to rethink the processor structure, abandoning the outdated physical separation between memory and computing units of von Neumann's architecture. Novel computing paradigms have emerged, such as in-memory computing (IMC), in which memory and computing are blended together in a single processing unit. IMC allows the complete elimination of the energy and latency overheads associated with the back-and-forth data transfer between memory and computing units. When implemented using crossbar arrays (CBAs) of memory devices, IMC represents a competitive candidate for next-generation energy-efficient accelerators for ML and AI on the edge. IMC was shown to be particularly effective in accelerating low-level, data-intensive algebraic operations such as matrix-vector (MVM) and inverse-matrix-vector multiplication (IMVM). However, while several IMC-based MVM demonstrators have been reported in recent years, IMVM demonstrators have faced additional challenges owing to the increased complexity of the circuit implementation, entailing analog feedback operation and suffering from increased sensitivity. Nonetheless, closed-loop IMC (CL-IMC) IMVM may capitalize on the IMC advantage even more than MVM, owing to the higher O(N^3) computational complexity, where N is the matrix size, which is instead reduced to O(1) in IMC-based systems. Here, we present a fully integrated IMC chip for IMVM designed and fabricated in 90~nm complementary metal-oxide-semiconductor (CMOS) technology. The chip features two 64×64 memory arrays, enclosed in an analog feedback loop by on-chip operational amplifiers (OAs), digital/analog (DAC), and analog/digital converters (ADC), providing the first complete primitive for acceleration of inverse operations under the IMC framework. We validate the integrated circuit (IC) by performing experiments on three real-life toy problems, namely, inversion of large-scale linear systems up to 512$\times$512 by recursive block inversion (RBI), sensor fusion by Kalman filter for trajectory estimation in sounding rockets, and acceleration of inverse kinematics in robotic arms for industrial automation and autonomous robots. Experimental results closely match the accuracy of fully digital systems working at the equivalent IC precision while simultaneously providing consistent advantages in terms of latency, energy, and area consumption. The obtained results represent the first large-scale experimental demonstration of the CL-IMC concept and consolidate its position as a promising candidate for next-generation energy-efficient accelerators on-the-edge.
- 16 December 2024
- Published by Springer Nature in Research Square
Open AccessAn SRAM-based fully-integrated analog closed-loop in-memory computing accelerator
Conventional processors are unsuited for the increasingly data-intensive workloads brought upon by the resurgence of artificial intelligence (AI) and machine learning (ML) in the latest years. The widespread diffusion of ML techniques in many different technological fields has prompted both academia and industry to rethink the processor structure, abandoning the outdated physical separation between memory and computing units of von Neumann's architecture. Novel computing paradigms have emerged, such as in-memory computing (IMC), in which memory and computing are blended together in a single processing unit. IMC allows the complete elimination of the energy and latency overheads associated with the back-and-forth data transfer between memory and computing units. When implemented using crossbar arrays (CBAs) of memory devices, IMC represents a competitive candidate for next-generation energy-efficient accelerators for ML and AI on the edge. IMC was shown to be particularly effective in accelerating low-level, data-intensive algebraic operations such as matrix-vector (MVM) and inverse-matrix-vector multiplication (IMVM). However, while several IMC-based MVM demonstrators have been reported in recent years, IMVM demonstrators have faced additional challenges owing to the increased complexity of the circuit implementation, entailing analog feedback operation and suffering from increased sensitivity. Nonetheless, closed-loop IMC (CL-IMC) IMVM may capitalize on the IMC advantage even more than MVM, owing to the higher O(N^3) computational complexity, where N is the matrix size, which is instead reduced to O(1) in IMC-based systems. Here, we present a fully integrated IMC chip for IMVM designed and fabricated in 90~nm complementary metal-oxide-semiconductor (CMOS) technology. The chip features two 64×64 memory arrays, enclosed in an analog feedback loop by on-chip operational amplifiers (OAs), digital/analog (DAC), and analog/digital converters (ADC), providing the first complete primitive for acceleration of inverse operations under the IMC framework. We validate the integrated circuit (IC) by performing experiments on three real-life toy problems, namely, inversion of large-scale linear systems up to 512$\times$512 by recursive block inversion (RBI), sensor fusion by Kalman filter for trajectory estimation in sounding rockets, and acceleration of inverse kinematics in robotic arms for industrial automation and autonomous robots. Experimental results closely match the accuracy of fully digital systems working at the equivalent IC precision while simultaneously providing consistent advantages in terms of latency, energy, and area consumption. The obtained results represent the first large-scale experimental demonstration of the CL-IMC concept and consolidate its position as a promising candidate for next-generation energy-efficient accelerators on-the-edge.
- Top Cited PapersOpen Access
- 1 October 2024
- Published by AIP Publishing in APL Materials
- Vol. 12 (10)
Roadmap to neuromorphic computing with emerging technologies
- 1 October 2024
- Published by AIP Publishing in APL Materials
- Vol. 12 (10)
Top Cited PapersOpen AccessRoadmap to neuromorphic computing with emerging technologies
- 6 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 71 (3) , 1900-1906
Compact Modeling and Mitigation of Parasitics in Crosspoint Accelerators of Neural Networks
In-memory computing (IMC) can accelerate data-intensive tasks, such as matrix-vector multiplication (MVM) or artificial neural networks (ANNs) inference, by means of the crosspoint memory array, allowing to reduce time and energy consumption. IMC accuracy, however, is affected by nonidealities, such as variability of the conductive weights or IR drop along wires due to parasitic resistances, whose impact steeply increases with the increase of array size. This work proposes a compact model to assess the impact of nonidealities for various circuital implementations, together with architectural schemes for their mitigation based on replicated arrays. The proposed mitigation techniques allow to restore the ANN accuracy from 72.7% to 94.9%, close to the software accuracy of 96.9%, in view of an increased area and energy consumption.
- 6 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 71 (3) , 1900-1906
Compact Modeling and Mitigation of Parasitics in Crosspoint Accelerators of Neural Networks
In-memory computing (IMC) can accelerate data-intensive tasks, such as matrix-vector multiplication (MVM) or artificial neural networks (ANNs) inference, by means of the crosspoint memory array, allowing to reduce time and energy consumption. IMC accuracy, however, is affected by nonidealities, such as variability of the conductive weights or IR drop along wires due to parasitic resistances, whose impact steeply increases with the increase of array size. This work proposes a compact model to assess the impact of nonidealities for various circuital implementations, together with architectural schemes for their mitigation based on replicated arrays. The proposed mitigation techniques allow to restore the ANN accuracy from 72.7% to 94.9%, close to the software accuracy of 96.9%, in view of an increased area and energy consumption.
- Open Access
- 21 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 71 (4) , 1839-1843
In-Memory Principal Component Analysis by Analogue Closed-Loop Eigendecomposition
Machine learning (ML) techniques such as principal component analysis (PCA) have become pivotal in enabling efficient processing of big data in an increasing number of applications. However, the data-intensive computation in PCA causes large energy consumption in conventional von Neumann computers. In-memory computing (IMC) significantly improves throughput and energy efficiency by eliminating the physical separation between memory and processing units. Here, we present a novel closed-loop IMC circuit to compute real eigenvalues and eigenvectors of a target matrix allowing IMC-based acceleration of PCA. We benchmark its performance against a commercial GPU, achieving comparable accuracy and throughput while simultaneously securing ×104 energy and ×102÷4 area efficiency improvements. These results support IMC as a leading candidate architecture for energy-efficient ML accelerators.
- 21 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 71 (4) , 1839-1843
Open AccessIn-Memory Principal Component Analysis by Analogue Closed-Loop Eigendecomposition
Machine learning (ML) techniques such as principal component analysis (PCA) have become pivotal in enabling efficient processing of big data in an increasing number of applications. However, the data-intensive computation in PCA causes large energy consumption in conventional von Neumann computers. In-memory computing (IMC) significantly improves throughput and energy efficiency by eliminating the physical separation between memory and processing units. Here, we present a novel closed-loop IMC circuit to compute real eigenvalues and eigenvectors of a target matrix allowing IMC-based acceleration of PCA. We benchmark its performance against a commercial GPU, achieving comparable accuracy and throughput while simultaneously securing ×104 energy and ×102÷4 area efficiency improvements. These results support IMC as a leading candidate architecture for energy-efficient ML accelerators.