Design and Applications of Emerging Computer

Systems 1st Edition Weiqiang Liu download

https://ebookmeta.com/product/design-and-applications-of-
emerging-computer-systems-1st-edition-weiqiang-liu/

Download more ebook from https://ebookmeta.com

We believe these products will be a great fit for you. Click
the link to download now, or visit ebookmeta.com
to discover even more!

Design of Semiconductor QCA Systems 1st Edition

Weiqiang Liu

https://ebookmeta.com/product/design-of-semiconductor-qca-
systems-1st-edition-weiqiang-liu/

Approximate Computing Weiqiang Liu

https://ebookmeta.com/product/approximate-computing-weiqiang-liu/

Computer Systems Digital Design Fundamentals of

Computer Architecture and ARM Assembly Language 2nd
Edition Ata Elahi

https://ebookmeta.com/product/computer-systems-digital-design-
fundamentals-of-computer-architecture-and-arm-assembly-
language-2nd-edition-ata-elahi-2/

Pelican 1st Edition Barbara Allen

https://ebookmeta.com/product/pelican-1st-edition-barbara-allen/
Tokyo Revengers Vol 5 1st Edition Ken Wakui

https://ebookmeta.com/product/tokyo-revengers-vol-5-1st-edition-
ken-wakui/

Bates Guide to Physical Examination and History Taking

12th Edition Lynn S Bickley Peter G Szilagyi

https://ebookmeta.com/product/bates-guide-to-physical-
examination-and-history-taking-12th-edition-lynn-s-bickley-peter-
g-szilagyi/

The Future of Global Retail Learning from China s

Retail Revolution 1st Edition Winter Nie

https://ebookmeta.com/product/the-future-of-global-retail-
learning-from-china-s-retail-revolution-1st-edition-winter-nie/

Her Dad’s Best Friend (The Man For Her) 1st Edition
Kate Hunt

https://ebookmeta.com/product/her-dads-best-friend-the-man-for-
her-1st-edition-kate-hunt/

C# 12 and .NET 8 Modern Cross-Platform Development

Fundamentals 8th Edition Mark J. Price

https://ebookmeta.com/product/c-12-and-net-8-modern-cross-
platform-development-fundamentals-8th-edition-mark-j-price/
Deconstructing Dr Strangelove The Secret History of
Nuclear War Films 1st Edition Sean M. Maloney

https://ebookmeta.com/product/deconstructing-dr-strangelove-the-
secret-history-of-nuclear-war-films-1st-edition-sean-m-maloney/
Weiqiang Liu
Jie Han
Fabrizio Lombardi Editors

Design and
Applications
of Emerging
Computer
Systems
Design and Applications of Emerging Computer
Systems
Weiqiang Liu • Jie Han • Fabrizio Lombardi
Editors

Design and Applications

of Emerging Computer
Systems
Editors
Weiqiang Liu Jie Han
College of Electronic and Information Department of Electrical and Computer
Engineering Engineering
Nanjing University of Aeronautics and University of Alberta
Astronautics Edmonton, AB, Canada
Nanjing, Jiangsu, China

Fabrizio Lombardi
Department of Electrical and Computer
Engineering
Northeastern University
Boston, MA, USA

ISBN 978-3-031-42477-9 ISBN 978-3-031-42478-6 (eBook)

https://doi.org/10.1007/978-3-031-42478-6

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland
AG 2024
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Paper in this product is recyclable.

Preface

Computing has a very different landscape in the era of artificial intelligence (AI).
Conventionally, high performance, low power, and uncompromising accuracy are
the main pillars of a computing system. With the rise of learning and data-intensive
applications, computing is faced with unprecedented and significant challenges that
deviate from those that can be coped with by using conventional methodologies.
As Dennard’s law is coming to an end, reduction in on-chip power consumption
and improvement in throughput due to technology scaling encounter serious
challenges; workloads of today’s applications (such as for AI, Big Data, and
Internet of Things) have also reached extremely high levels of complexity in
computation. Power dissipation has become one of the fundamental barrier to
scale computing performance across all technology platforms. Therefore it is no
surprise that computation at nanoscale requires innovative approaches. Moreover,
many so-called emerging computing paradigms have been widely studied (e.g.,
probabilistic, stochastic, neuromorphic, in-memory), mostly at system level to
alleviate these encountered hurdles; however, their successful evolution necessitates
implementations that require efficient circuits in a multitude of modules (such as
memory, arithmetic, and control). Substantial challenges remain also at architectural
and system levels. Although bridging of technology with design has attracted
significant attention from academic and industrial communities in the past decade,
it still requires considerable efforts to accomplish implementations that are energy-
efficient and high-performance for systems in so many diverse applications.
This book addresses the technological contributions and developments at various
hardware levels of new computing paradigms that include stochastic, probabilis-
tic, neuromorphic, spintronic, bio-inspired, in-memory computing, and quantum
computing. With this book, it is expected to collect state-of-the-art progress in
different emerging computing areas; this is achieved by chapters covering the entire
spectrum of research activities in such emerging computing paradigms, so bridging
the entire system stack, i.e., from devices, circuits, architectures, up to systems.
The book covers tutorials, reviews, and surveys of current theoretical/experimental
results, design methodologies, and applications over a wide spectrum of scope for
an enlarged and specialized readership.

v
vi Preface

Emerging computing paradigms have shown great potential for many appli-
cations; however, they are not yet at a fully mature stage. A comparison of
unconventional computing techniques in different dimensions is shown in Fig. 1;
hence this book provides a comprehensive reference for current developments and
promotes future research on this important research area from which innovative
computing schemes can be derived.
The chapters in this book are divided into four parts.
The first part presents in-memory computing for neuromorphic computing
and machine learning applications based on emerging magnetic devices [1]. A
bottleneck for high performance is the relatively slow communication between
memory and processors [2]. As a remedy to this issue, in-memory computing has
shown promising prospects and neuromorphic computing has provided a viable
means, especially for data-intensive applications such as machine learning [3]. By
exploiting the switching properties of magnetic materials and peripheral circuit
design, logical and arithmetic operations within the memory array are realized and
further applied to neural networks and machine learning, greatly alleviating the

Fig. 1 Comparison of unconventional computing techniques in different dimensions

Preface vii

“Energy Wall” and “Memory Wall” of traditional computing architectures [4]. Sev-
eral important approaches to achieving in-memory logic and arithmetic computing
(e.g., XOR, AND, and Multiply-and-Accumulate) [5], in-memory neuromorphic
computing [6], and other new in-memory applications [7] are discussed in Part I of
this book.
The second part introduces stochastic computing and its applications. Stochastic
computing utilizes random (and sometimes, deterministic) binary bit streams to
encode information [8]. Applications of stochastic computing include image and
digital signal processing [9]. Recently, it has been shown that it is effective for
the inference and training of neural networks [10]. Toward low power, stochastic
computing exploits the simplicity in statistical processing using simple logic
circuits; however, this benefit comes at the cost of a long latency due to the
long stochastic sequence length for achieving a high accuracy [11]. This tradeoff
between hardware efficiency and long latency has been addressed using various
approaches, including the design and characterization of efficient random number
generators, parallel stochastic computing multipliers, and morphological neural
networks. These are the topics addressed in Part II of this book.
The third part covers inexact or approximate computing from circuits to AI and
communication applications. Emerging computing applications exhibit character-
istics of error tolerance or resilience. The requirement of accuracy, therefore, is
relieved, especially in the intermediate stages of a computing process, for gains
in performance and power dissipation. The paradigm that exploits accuracy as a
major design consideration has been referred to as approximate computing [12].
Recently, various approximate arithmetic circuits under different design constraints
have been developed [13]. At the same time, substantial research has focused
on circuit level techniques for approximating logic and memory design [14].
Approximate computing has found important applications in arithmetic circuits,
wireless communication, machine learning accelerators, and neural networks. It
is worth mentioning that approximate computing can significantly improve the
efficiency of AI applications by exploiting its error-tolerant nature [15]. Part III of
this book discusses these important approximate computing approaches.
Finally, the last part includes quantum computing and other emerging com-
puting topics. Quantum computing relies on the superposition and entanglement
of physical states at the microscopic scale, so it provides ultimate performance
and efficiency by leveraging the quantum mechanical behavior of devices [16].
Various research directions have been pursued by building scalable physical systems
for quantum computing and investigating a number of emerging devices [17]. In
addition to the challenges in hardware, algorithms are an aspect that is crucial
for the scalability and robustness of quantum computers [18]. In contrast, the
Ising model provides a mathematical description of the electromagnetic interactions
among an array of spins [19]. It has been shown that computers built on the
principle of the Ising model, or Ising machines, are efficient solvers of combinatorial
viii Preface

optimization problems [20]. Finally, network-on-chips have been developed for

multi-core or multi-chip systems for efficient communication and computation [21].
The design and algorithm aspects of these emerging computing paradigms using
CMOS implementations and quantum dots are some of the additional subjects of
Part IV of this book.

Nanjing, Jiangsu, China Weiqiang Liu

Edmonton, AB, Canada Jie Han
Boston, MA, USA Fabrizio Lombardi

References

1. C. Chang et al., NV-BNN: an accurate deep convolutional neural network based

on binary STT-MRAM for adaptive AI edge, in 56th Annual Design Automation
Conference, (2019)
2. M. Horowitz, 1.1 Computing’s energy problem (and what we can do about it),
in IEEE International Solid-State Circuits Conference (ISSCC), San Francisco,
2014, pp. 10–14
3. S. Jung et al., A crossbar array of magnetoresistive memory devices for in-
memory computing. Nature 601(7892), 211–216 (2022)
4. S. Angizi, Z. He, A. Awad, D. Fan, MRIMA: An MRAM-based in-memory
accelerator. IEEE Trans. Comput. Aided Des. Integr. Circuits Syst. 39(5), 1123–
1136 (2020)
5. S. Jain et al., Computing-in-memory with spintronics, in 2018 Design, Automa-
tion & Test in Europe Conference & Exhibition (DATE), (2018)
6. L. Chang et al., CORN: In-buffer computing for binary neural network, in 2019
Design, Automation & Test in Europe Conference & Exhibition (DATE), (2019)
7. X. Wang et al., TCIM: Triangle counting acceleration with processing-in-
MRAM architecture, in 2020 57th ACM/IEEE Design Automation Conference
(DAC), (2020)
8. B.R. Gaines, Stochastic computing systems, in Advances in Information Sys-
tems Science, (Springer, New York, 1969), pp. 37–172
9. Alaghi, W. Qian, J.P. Hayes, The promise and challenge of stochastic comput-
ing. IEEE Trans. Comput. Aided Des. Integr. Circuits Syst. 37(8), 1515–1531
(2017)
10. Y. Liu, S. Liu, Y. Wang, F. Lombardi, J. Han, A survey of stochastic computing
neural networks for machine learning applications. IEEE Trans. Neural Netw.
Learn. Syst. 32(7), 2809–2824 (2021)
11. S. Liu, W.J. Gross, J. Han, Introduction to dynamic stochastic computing. IEEE
Circuits Syst. Mag. 20(3), 19–33 (2020)
12. W. Liu, F. Lombardi, M. Schulte, Approximate computing: From circuits to
applications. Proc. IEEE 108(12), 2103–2107 (2020)
Preface ix

13. H. Jiang, F.J.H. Santiago, H. Mo, L. Liu, J. Han, Approximate arithmetic cir-
cuits: A survey, characterization, and recent applications. Proc. IEEE 108(12),
2108–2135 (2020)
14. S. Amanollahi, M. Kamal, A. Afzali-Kusha, M. Pedram, Circuit-level tech-
niques for logic and memory blocks in approximate computing systems. Proc.
IEEE 108(12), 2150–2177 (2020)
15. S. Venkataramani et al., Efficient AI system design with cross-layer approxi-
mate computing. Proc. IEEE 108(12), 2232–2250 (2020)
16. H. Mooij, The road to quantum computing. Science 307(5713), 1210–1211
(2005)
17. J.E. Mooij, T.P. Orlando, L. Levitov, L. Tian, C.H. Van der Wal, S. Lloyd,
Josephson persistent-current qubit. Science 285(5430), 1036–1039 (1999)
18. T.D. Ladd, F. Jelezko, R. Laflamme, Y. Nakamura, C. Monroe, J.L. O’Brien,
Quantum computers. Nature 464(7285), 45–53 (2010)
19. Lucas, Ising formulations of many NP problems. Front. Phys., 5 (2014)
20. K. Yamamoto, K. Kawamura, K. Ando, N. Mertig, T. Takemoto, et al.,
STATICA: A 512-spin 0.25 m-weight annealing processor with an all-spin-
updates-at-once architecture for combinatorial optimization with complete
spin–spin interactions. JSSC 56(1), 165–178 (2020)
21. A.E. Kiasari, Z. Lu, A. Jantsch, An analytical latency model for networks-on-
chip. IEEE Trans. Very Large Scale Integr. VLSI Syst. 21(1), 113–123 (2013)
Contents

Part I In-Memory Computing, Neuromorphic Computing and

Machine Learning
Emerging Technologies for Memory-Centric Computing . . . . . . . . . . . . . . . . . . . 3
Paul-Antoine Matrangolo, Cédric Marchand, David Navarro,
Ian O’Connor, and Alberto Bosio
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Resistive Random Access Memory (RRAM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Memory Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 IMC Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3 Spin-Transfer Torque Magnetoresistive Random Access Memory . . . . . . . . . 11
3.1 Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Memory Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3 IMC Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4 Phase-Change Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.1 Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Memory Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3 IMC Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5 FeFET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.1 Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 Memory Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 IMC Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6 Comparison and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
An Overview of Computation-in-Memory (CIM) Architectures . . . . . . . . . . . 31
Anteneh Gebregiorgis, Hoang Anh Du Nguyen, Mottaqiallah Taouil,
Rajendra Bishnoi, Francky Catthoor, and Said Hamdioui
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2 Classification of Computer Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

xi
xii Contents

2.1 Classification Based on Computation Position . . . . . . . . . . . . . . . . . . . . . . . . 33

2.2 Classification Based on Memory Technology . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 Classification Based on Computation Parallelism . . . . . . . . . . . . . . . . . . . . . 36
3 Computation-in-Memory-Array (CIM-A) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1 DRISA-3T1C: A DRAM-Based Reconfigurable In Situ
Accelerator with 3 Transistors and 1 Capacitor (3T1C) Design . . . . . . 38
3.2 CRS: Complementary Resistive Switch Architecture . . . . . . . . . . . . . . . . . 41
3.3 CIM: Computation-in-Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 PLiM: Programmable Logic-in-Memory Computer . . . . . . . . . . . . . . . . . . 43
3.5 MPU: Memristive Memory Processing Unit . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.6 ReVAMP: ReRAM-Based VLIW Architecture . . . . . . . . . . . . . . . . . . . . . . . 46
4 Computation-in-Memory-Peripherals (CIM-P) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.1 S-AP: Cache Automaton. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2 ISAAC: A Convolutional Neural Network Accelerator with
In Situ Analog Arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3 PRIME: A Processing-in-Memory Architecture for Neural
Network Computation in ReRAM-Based Main Memory . . . . . . . . . . . . . 53
4.4 STT-CiM: Computing-in-Memory Spin-Transfer Torque
Magnetic RAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.5 DPP: Data Parallel Processor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.6 R-AP: Resistive RAM Automata Processor . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5 CIM Design-flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1 System-Level Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2 Circuit-Level Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Toward Spintronics Non-volatile Computing-in-Memory Architecture . . . 67
Bi Wu, Haonan Zhu, Tianyang Yu, and Weiqiang Liu
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2 MRAM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3 Implementation of Boolean Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.1 Analog Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.2 Read/Write Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.3 Cell Modification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.4 Peripheral Circuit Modification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4 MRAM for Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.1 CNN and BNN Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2 MRAM for XNOR Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.3 MRAM for Bit-Count Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.4 MRAM for Max-Pool Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.5 MRAM for BNN Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.6 MRAM for Analog Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5 MRAM for Other Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.1 DNA Read Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Contents xiii

5.2 Triangle Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

5.3 True Random Count Generators. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Is Neuromorphic Computing the Key to Power-Efficient Neural
Networks: A Survey. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Muhammad Hamis Haider, Hao Zhang, S. Deivalaskhmi,
G. Lakshmi Narayanan, and Seok-Bum Ko
1 Introduction to Biomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2 Understanding the Biological Neuron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.1 The Response to Extra-cellular Stimulus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3 Neuromorphic Computing: Learning from the Brain. . . . . . . . . . . . . . . . . . . . . . . . 97
4 Spiking Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.1 Spiking Neuron Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.2 Common Training Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5 Hardware Accelerators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.1 General Mesh Accelerators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.2 Feedforward Accelerators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.3 Resource Analysis and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6 Limitations and Potential Improvements of SNNs . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Emerging Machine Learning Using Siamese and Triplet
Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Ziheng Wang, Farzad Niknia, Shanshan Liu, Pedro Reviriego,
and Fabrizio Lombardi
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
2 Multi-Branch NNs: Algorithms and Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2.1 Siamese Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2.2 Triplet Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2.3 ASIC-Based Design for a Branch Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
3 Error Tolerance of Multi-Branch NNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.1 Error Tolerance During Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.2 Error Tolerance During Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
An Active Storage System for Intelligent Data Analysis
and Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Shengwen Liang, Ying Wang, Lei Dai, Yingying Chen, Renhai Chen,
Fan Zhang, Gong Zhang, Huawei Li, and Xiaowei Li
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
2 Background and Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
2.1 Unstructured Data Analysis System. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
xiv Contents

2.2 Near Data Processing and Deep Learning Accelerator. . . . . . . . . . . . . . . . 150

2.3 Learned Data Cache and Placement. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
2.4 Active Store . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
3 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
3.1 Execution Time Breakdown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
3.2 ANNS Algorithm Exploration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4 Active Storage System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.1 The Active Storage Software: DHS Library . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.2 Hardware Architecture: The Active Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.3 The Procedure of Data Retrieval in Active Storage . . . . . . . . . . . . . . . . . . . 159
5 DHS-x Accelerator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.1 Architecture: Direct Flash Accessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.2 I/O Path in Active Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.3 Hybrid Search Engine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.4 LSTM-Based Data Cache and Placement. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.1 Hardware Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.3 Evaluation of DHS Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
6.4 Evaluation of DHS-x Accelerator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.5 The Single-Node System Based on Active Storage . . . . . . . . . . . . . . . . . . . 175
6.6 Data Cache and Placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Error-Tolerant Techniques for Classifiers Beyond Neural
Networks for Dependable Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Shanshan Liu, Pedro Reviriego, Xiaochen Tang, and Fabrizio Lombardi
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
2 K Nearest Neighbors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
2.1 Errors in KNNs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
2.2 Voting-Margin-Based Error-Tolerant KNNs for Binary Classification 189
2.3 K + 1 Nearest Neighbors for Multiclass Classification . . . . . . . . . . . . . . . 191
3 Ensemble Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
3.1 Random Forest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
3.2 Voting-Margin-Based Error-Tolerant Random Forest . . . . . . . . . . . . . . . . . 196
4 Support Vector Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
4.1 SVM with Different Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
4.2 Result-Based Re-Computation Error-Tolerant SVM . . . . . . . . . . . . . . . . . . 200
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Contents xv

Part II Stochastic Computing

Efficient Random Number Sources Based on D Flip-Flops for
Stochastic Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Kuncai Zhong and Weikang Qian
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
2 Guidelines for Applying LFSRs in Stochastic Computing . . . . . . . . . . . . . . . . . . 214
2.1 Model and Evaluation Methodology for LFSRs . . . . . . . . . . . . . . . . . . . . . . . 214
2.2 The Selection of the Feedback Polynomials for LFSR Pair. . . . . . . . . . . 215
2.3 The Selection of the Seeds for Identical-Feedback LFSR Pair . . . . . . . 216
3 Proposed Method for Building Successive RNSs. . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
3.1 Method for Building Two Successive RNSs . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
3.2 Method for Building Multiple Successive RNSs . . . . . . . . . . . . . . . . . . . . . . 220
4 Method for Building Non-successive RNSs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
5 Efficient Designs of RNSs Based on DFFs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
6 Experimental Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
6.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
6.2 Accuracy Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.3 Area Comparison. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
6.4 Area–Accuracy Trade-off Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
Stochastic Multipliers: from Serial to Parallel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Yongqiang Zhang, Jie Han, and Guangjun Xie
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
2.1 Binary Multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
2.2 Booth Multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
2.3 Stochastic Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
2.4 SC Correlation and Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
3 The Design Approaches of Stochastic Multipliers. . . . . . . . . . . . . . . . . . . . . . . . . . . 244
3.1 Stochastic Multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
3.2 Exact Stochastic Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
Applications of Ising Models Based on Stochastic Computing . . . . . . . . . . . . . . 265
Duckgyu Shin, Naoya Onizawa, Warren J. Gross, and Takahiro Hanyu
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
2 Preliminaries. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
2.1 SA for the Ising Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
2.2 P-bit-Based Ising Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
3 Stochastic Simulated Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
3.1 Spin Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
3.2 Annealing Process of SSA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
xvi Contents

4 CMOS Invertible Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272

4.1 Basics of CIL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
4.2 Training BNNs Based on CIL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
5 CIL Training Hardware Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
5.1 Architecture of CIL Training Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
5.2 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Stochastic and Approximate Computing for Deep Learning: A Survey . . . 281
Tina Masoudi, Hao Zhang, Aravindhan Alagarsamy, Jie Han,
and Seok-Bum Ko
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
2.1 Stochastic Computing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
2.2 Approximate Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
2.3 Deep Learning Arithmetic Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
2.4 Deep Learning Application Accelerators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
3 Stochastic and Approximate Computing-Based Deep Learning
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
4 Area- and Power-Efficient Hybrid Stochastic-Approximate
Designs in Deep Learning Applications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
4.1 Low Complexity with High-Accuracy Winograd
Convolution Based on Stochastic and Approximate Computing . . . . . 295
4.2 A Hybrid Power- and Area-Efficient
Stochastic-Approximate Neuron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
6 Future Research Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
Stochastic Computing Applications to Artificial Neural Networks . . . . . . . . . 303
Josep L. Rosselló, Joan Font-Rosselló, Christiam F. Frasser,
Alejandro Morán, Vincent Canals, and Miquel Roca
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
2 Stochastic Computing Basic Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
2.1 Stochastic Signals and Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
3 Classical Artificial Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
3.1 Application of Stochastic Computing to Artificial Neural Networks. 309
4 Morphological Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
4.1 Application of Stochastic Computing to the Implementation
of Morphological Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
Exploring the Variety of Random
Documents with Different Content
The Project Gutenberg eBook of Laboratory
This ebook is for the use of anyone anywhere in the United States
and most other parts of the world at no cost and with almost no
restrictions whatsoever. You may copy it, give it away or re-use it
under the terms of the Project Gutenberg License included with this
ebook or online at www.gutenberg.org. If you are not located in the
United States, you will have to check the laws of the country where
you are located before using this eBook.

Title: Laboratory

Author: Jerome Bixby

Illustrator: Ed Emshwiller

Release date: May 10, 2019 [eBook #59470]

Language: English

Credits: Produced by Greg Weeks, Mary Meehan and the Online

Distributed Proofreading Team at http://www.pgdp.net

*** START OF THE PROJECT GUTENBERG EBOOK LABORATORY ***

LABORATORY
 BY JEROME BIXBY

Trying to keep a supercolossal laboratory

invisible when two curious aliens are poking
around can be a trying affair for even the
most brilliant of minds....

[Transcriber's Note: This etext was produced from

Worlds of If Science Fiction, December 1955.
Extensive research did not uncover any evidence that
the U.S. copyright on this publication was renewed.]
Gop's thoughts had the bluish-purple tint of abject apology: "They're
landing, Master."
Pud looked up from the tiny thig-field he had been shaping in his
tentacles. "Of course they are," he thought-snapped. "You practically
invited them down, didn't you? If you'd only kept a few eyes on the
Detector, instead of day-dreaming—"
"I'm sorry," Gop said unhappily. "I wasn't day-dreaming, I was
observing the magnificent skill and finesse with which you shaped
the thig. After all, this system is so isolated. No one ever came along
before.... I just supposed no one ever would—"
"A Scientist isn't supposed to suppose! Until he's proven wrong, he's
supposed to know!" Thirty of Pud's eyes glowered upward at the tiny
alien spaceship, only ninety or so miles above the surface of the
laboratory-planet and lowering rapidly. The rest of Pud's eyes—more
than a hundred of them, set haphazardously in his various-sized
heads like gurf-seeds on rolls—scoured every inch of the planet's
visible surface, to make certain that no sign of the Vegans' presence
on the planet, from the tiniest experiment to the gigantic servo-
mechanical eating pits, was left operating or visible.
Irritatedly he squelched out of existence a yim-field that had taken
three weeks of laborious psycho-induction to develop. His psycho-
kineticut stripped it of cohesion, and its faint whine-and-crackle
vanished.
"I told you to deactivate all our experiments," he snapped at Gop.
"Don't you understand Vegan?"
Abashed, the Junior Scientist lowered his many eyes.
"I—I'm sorry," Gop said humbly. "I thought the yim might wait until
the creatures landed, Master ... perhaps their auditory apparatus
would not have been sufficient to reveal its presence to them, in
which case the field would not have had to be—"
"All right, all right," Pud grunted. "I appreciate your point ... but,
dripping mouthfuls, you know that any risk of detection is too great.
You know the regulations on Contact!"
"Yes, Master."
"Speaking of which, part of your seventh head is showing."
The Junior Scientist included the head in the personal invisibility field
which he himself was broadcasting.
"Of all the suns in this sector," Pud thought, eying the little
spaceship, "and of all the planets around this particular sun, they
have to choose this one to land on. Chew!"
Gop flushed. A member of the Transverse Colon Revivalists, he
found Pud's constant atheistic swearing very disturbing. He sighed
inwardly. Usually at least one of Pud's heads could manage to keep
its sense of humor, but right now all of them were like proton-
storms. The Senior Scientist was on the verge of one of his
totalitantrums.
"They must have sighted flashes from our experiments," Pud went
on, "before you decided you could spare just one set of eyes for the
Detector!"

Though both Vegans were invisible to other eyes, they remained

visible to each other because their eyes were adjusted to the
wavelength of their invisibility fields. By the same token, they could
see all their invisible experiments—a vast litter of gadgets, gismos,
gargantuan gimmicks, shining tools, huge and infinitesimal
instruments, stacks of supplies, and various types of energy fields,
the latter all frozen in mid-activity like smudges on a pane of glass.
The sandy ground was the floor of the Vegans' laboratory; small hills
and outcroppings of rock were their chairs and work-benches. Like a
spaceship junkyard, or an enormous open-air machinery warehouse,
the laboratory stretched away from the two Scientists in every
direction to the planetoid's near horizon.
Pud intensified the general invisibility field to the last notch, and the
invisible experiments became even more invisible.
The thig-field was a nameless-colored whorl of energy in the Senior
Scientist's tentacles. In his concern for the other experiments, he
had forgotten to deactivate it. It grew eagerly to the size of a back
yard, then of a baseball diamond, then of a traffic oval, and one
shimmering edge of it touched his body, which he had not insulated.
Energy crackled. Pud jumped forty feet into the air, swearing, and
slapped the field into non-existence between two tentacles.
His body, big as an apartment house, floated slowly downward in the
laboratory-planet's light gravity.
The tiny alien spaceship touched the ground just as he did. The
rocket flare flickered and died.
The ship sat on its fins, about thirty feet—Vegan feet—away. In its
shining side, a few Vegan inches above the still smoking rocket
tubes, was a small black hole.
"Master, look!" Gop thought. "Their ship is damaged ... perhaps
that's why they landed!" And he started to extend a tentative extra-
sensory probe through the hole.
Pud lashed out with a probe of his own, knocking Gop's aside before
it could enter the hole. "Nincompoop! ... don't go esprobing until we
know if they're sensitive to it or not! Can't you remember the
regulations on Contact for just one minute?"
The tiny spaceship sat silently, while its occupants evidently studied
the lay of the land. Small turrets halfway up its sides twitched this
way and that, pointing popgun armament.
Pud inspected the weapons extra-sensorily, and thought an amused
snort: the things tossed a simple hydrogen-helium pellet for a short
distance.
Gop, nursing a walloping headache as a result of Pud's rough
counterprobe, thought sourly to himself: "I try to save the yim ...
that's wrong. He forgets to deactivate the thig ... that's all right. I
esprobe ... that's wrong. He esprobes ... that's all right."
At last: "They're getting out," Gop observed.
A tiny airlock had opened in the side of the ship. A metal ladder
poked out, swung down, settled against the ground.
The aliens—two of them—appeared; looked down, looked up, looked
to the right and to the left. Then they came warily down the ladder.
For a few minutes the giant Vegans watched the creatures wander
about. One of them approached one of Pud's tails. Irritatedly Pud
lifted it out of the way. The little creature snooped on, unaware that
twenty tons of invisible silicoid flesh hung over its head. Pud curled
the tail close to him, and did likewise with all his other tails.
"You'd better do the same," he advised Gop, his thought-tone
peevish.
Silently, Gop drew in his tails. One unwise move, he knew, and the
Senior Scientist would start thinking in roars.
One of Gop's tails scraped slightly against a huge boulder. The scales
made a tractor-on-gravel sound.
Pud thought in roars.
The tiny creature had stopped and was turning its helmeted head
this way and that, as if trying to see where the sound had come
from. It had drawn a weapon of some sort from a holster at its belt
—another thermonuclear popgun.
The creature turned and came back toward the Vegans, heading for
his ship. Pud lifted his tail again. The creature passed under it,
reached the ship, joined its partner.

"I heard it too, Johnny," Helen Gorman said nervously. "A loud
scraping noise—"
"It seemed to come from right behind me," Johnny Gorman said.
"Damn near scared me off the planet ... I thought it was a rockslide.
Or the biggest critter in creation, sneaking up on me. I couldn't see
anything, though ... could you?"
"No."
Johnny stood there, blaster in hand, looking around, eyes sharp
behind his faceplate. He saw nothing but flat, grayish-red ground, a
scattering of stone outcroppings large and small; nothing but the
star-clouded black of space above the near horizon, and the small
sun of the system riding a low hillock like a beacon.
"Blue light," he said thoughtfully. "Green light. Red and purple lights.
And a mess of crazy colors we never saw before. Whatever those
flashes were, honey, they looked artificial to me...."
Helen frowned. "We were pretty far off-world when we saw them,
Johnny. Maybe they were aurorae—or reflections from mineral
pockets. Or magnetic phenomena of some kind ... that could be why
the ship didn't handle right during landing—"
Johnny studied the upside-down dials on the protruding chest-board
of his spacesuit.
"No neon in the atmosphere," he said. "Darned little argon, or any
other inert gas. The only large mineral deposits within fifty miles are
straight down. And this clod's about as magnetic as an onion." He
gave the surrounding bleak terrain another narrow-eyed scrutiny. "I
suppose it could have been some kind of aurora, though ... it's gone
now, and there isn't a sign of anything that could have produced
such a rumpus." He looked around again, then sighed and finally
holstered his blaster. "Guess I'm the worrying type, hon. Nothing
alive around here."
"I wonder what that sound was."
"Probably a rock falling. This area's been undisturbed for God knows
how many million years ... the jolt of our landing just shook things
up a little." He grinned, a little sheepishly. "As for the landing ... I
was so scared after that meteor hit us, it's a wonder I didn't nail the
ship halfway into the planet, instead of just jolting us up."
Helen looked up at the three-foot hole in the side of the ship.
Johnny followed her gaze, and grunted. "We'd better get to work."
He turned to the ladder that led up to the airlock. "I'll rig the
compressor to charge the spare oxy-tanks ... we'll have to delouse
this air of ammonia, but otherwise it's fine. Look, honey, I won't
need any help; why don't you get busy on a PC?"
Helen nodded, still staring up at the meteor-hole. "You know," she
said slowly, "it wouldn't happen again this way in a million years,
Johnny. Thank God, this clod was here ... we ought to name it
Lifesaver."
"Yeah, sure," Johnny said ironically. "It'll save our lives. Only thing is,
it got us into this mess in the first place!"
He started up the ladder, using only his arms, legs trailing.
Helen got down on hands and knees and began poking around for
the two dozen or so samples needed for Standard Planetary
Classification. Bits of rock, air, vegetable growth, dust—the dust was
very important. All went into vac-containers at her belt.
Then suddenly she said, "O-o-o-oof!" and reared back on her knees
and clapped both hands to her helmet. Her eyes squeezed shut
behind her faceplate, then opened wide and frightened.
By the time her hands reached her helmet, Johnny had his blaster
out and was floating toward the ground, looking around for
something to shoot at. His boots touched, and two long light-gravity
steps brought him to her side.

Pud had been leaning over the tiny spaceship, one of his faces only
feet above the little creatures.
Gop's thought came: "What are they?"
"Fanged if I know. Bipeds ... never saw such little ones." Pud
adjusted several eyes to a certain wavelength and studied the
creatures through their spacesuits. He gave Gop a thought-nod:
"Mammals. Bi-sexual. They're probably mates."
"It's a miracle they didn't land right in the middle of one of our
experiments."
That brought back Pud's ill-temper. "Miracle! Didn't you see me give
this cosmic kiddycar of theirs a couple of psychokineticlouts so
they'd land where they did?" The Senior Scientist glared around at
their thousand-and-one experiments, and then down at the little
spaceship, smaller than the smallest of them, squatting on toy fins.
He curled a tentacle, as if wishing he could swat it.
Gop knew, however, that despite Pud's irritation at having his work
interrupted, he was just a little intrigued by the aliens. No matter
how insignificant they were they were animate life of some
intelligence, and Pud must be wondering about them.
Gop thought it might be a good idea to dwell on that, in order to
keep Pud from getting his heads in an uproar again.
"Can you get into their thoughts?" he inquired.
"I haven't tried. I don't think I could keep my potential down to their
level."
"Wonder where they're from."
"Who cares?" Pud snorted. "I just wish they'd go away."
Gop noted, though, that Pud's heads were lowering closer over the
creatures.
"They're nowhere near acceptable Contact level, are they?" Gop
said, after a moment.
"From their appearance, I'd say they're even beneath classification.
Reaction motor in their ship. Primitive weapons. Protective garments
... they can't even adjust physically to hostile environments!"
A minute passed.
Pud said, "Mm. Well. I think I will see what I can read ... just to
have something to talk about at the Scientists' Club."
He sent out a tentative probe ... a little one ... just enough to
register in one of his brains the total conscious content of one of the
little creature's minds. He was afraid to go deeper, after the
subconscious, though actually that was far more important. But deep
probing would probably be felt for what it was, while conscious
probing was just a little painful.
The creature popped erect in its squatting position, and clapped its
upper extremities to its head.
The other one, which had been scrambling up the ladder to the
ship's airlock, drew its popgun and joined the first.
"They're from someplace called Earth," Pud said. "In the V-LM-12Xva
Sector of this Galaxy, as nearly as I can make out. They're an
Exploration Team, sent out by their planet to gather data on the
nature of the physical universe." He paused to consult the third
memory bank of his fifth brain, where he had impressed the content
of the creature's mind. "They've had space travel for about two
hundred of their years. I translate that as about eleven of ours." He
consulted again. "Highly materialistic. Externally focused. Very
limited sensorium. An infant race, chasing everything that moves,
round and round through their little three-dimensional universe.
They've a long way to go."
"What are they doing here?"
"Hm." Pud consulted again. "A routine exploration flight brought
them to this system ... and an almost unbelievable coincidence has
served to delay them here. They dropped their meteor-screens for
just a moment—at just the wrong moment. A large meteor came
along, entered the ship, and destroyed both their atmosphere-
manufacturing equipment and the large pressure tank of atmosphere
which they kept as reserve in case the equipment should fail." He
paused. "Mixture of hydrogen and oxygen ... they can't live without
it. At any rate, the ship was evacuated, and they barely had time to
get into the ... mm, spacesuits, they call them ... which they now
wear. The accident left them with no atmosphere whatever, except
the small amount in the tanks of those suits. That will be exhausted
in a short time ... I gather that if this planet hadn't been here, they'd
have been goners. As it stands, they plan to charge their spare suit-
tanks, which weren't harmed, with the air of this planet, and then
return to their Earth, subsisting on the tanked air, by hyperspatial
drive...." Again Pud paused. "Hm. Well, now! I'd overlooked that. So
they have hyperspatial drive, at least ... and after only two hundred
years of space travel! Hm. Perhaps they are worth a closer look...."
Pud lowered his heads over the two little aliens, who were moving
warily, popguns drawn, away from the ship.
"Pud," Gop said nervously.
"What?"
"One of them is crawling toward the time-warp."
"Well, don't tell me about it ... lift the warp out of the way!"
Gop extended a tentacle, first reconstituting it on the seventh atomic
sublevel so he wouldn't get it blown off, and gently picked up the
time-warp. It looked like a blue-violet frozen haze in his grasp. He
set it down on the other side of the spaceship, anchoring it again to
now so it wouldn't go flapping off along the time-continuum.
"So they didn't land because they saw flashes from our
experiments," he said a little triumphantly.
One of Pud's heads turned and gave the Junior Scientist an acid
look, while the others continued to observe the aliens.
"They lowered their meteor-screens," he said nastily, "thus bringing
about this entire bother, because they wanted to get a better look at
the flashes."
Gop was silent, but he thought acidly: "That's what you say—you
won't let me esprobe, and when you do, you manage to prove it's all
my fault."

Johnny Gorman had just said to Helen, "I want to chip a few
samples off that outcropping over there ... come on, hon."
He started toward the ridge of gray-black rock. Helen followed on his
heels.
"As-pir-in," she said, deliberately falsetto, and her helmet-valet fed
her another pill with a sip of water.
"Then we'll go back and stick inside the ship until the tanks are
charged," Johnny went on, a little grimly. "I think we're just edgy.
Planets don't give people headaches ... and there's nothing alive
within in a million miles of this dustball." He hefted his blaster, which
he had adjusted to Wide-Field. "But just in case...."

"Pud," Gop said, still more nervously.

"Yes, I see, you idiot! Lift the tharn-field out of their way ... I'll take
care of the space-warp generator!"
The giant Vegans, for all their bulk, moved soundlessly and at great
speed until they were between the aliens and the stone outcropping
toward which they appeared to be heading. Gop extended a
tentacle, curled it at an odd angle, and picked up the shimmering
tharn-field, which was the Vegans' reservoir of Basic Universal
Energy. Set in any energy matrix, tharn became that energy; added
to any existing energy, tharn augmented it to any desired potential.
Thus it was extremely valuable to their experiments ... and very risky
stuff to handle, as well.
Gingerly, Gop set the tharn down beyond the outcropping. At the
same time he picked up several instruments that lay nearby—an
electron-wrench, a snurling-iron, a plotz-meter, several pencil-rays.
He placed them on the ground beside the tharn.
Pud had curled twelve tentacles around the space-warp generator—
it was as big as a city block, and heavy, even in light gravity. He
puffed a thought at Gop: "Give me a tentacle."
Gop helped his Master place the generator safely on the other side
of the ridge.

Johnny Gorman banged off a handful of rock, and shoved it into the
vac-container at his belt.
"Okay, hon," he said. "Let's go."
They stood once more moment atop the ridge, looking out over the
barren, rusty-gray plain that the ridge had until now concealed from
their gaze.
"Looks just as dead as the rest," Johnny observed. "I guess we were
just jumpy over nothing." He turned to start down the slope. "Come
on."
In three long light-gravity steps he had reached the bottom, and
turned to steady Helen.
She wasn't there.
She had tripped and tumbled off the other side of the ridge. He
could hear her screaming.

"Putrefied proteins!" Pud roared. "Help me get it out of the tharn!"

The two Vegans leaned over the ridge. While Gop forced the
writhing folds of the tharn-field apart with two reconstituted
tentacles, Pud reached in, plucked the little alien out and set it
upright.
It immediately scrabbled up the side of the ridge as fast as it could
and joined its mate, which had bounded up the other side.
"Now look at what you've done!" Pud raged. "What about the rules
on Contact! The Examiners will get this out of us when we report on
our Projects ... mountains of bites, we've revealed ourselves!"
"Not really, Master," Gop said, rushing his thoughts. "All the creature
will know is that it tumbled into the field, and then was somehow
ejected by it ... a trick of gravity, perhaps ... a magnetic vortex ... it
won't know what really happened—"
"That—field—was—supposed—to—be—turned—off," Pud said, every
one of his faces green with rage.
"I—"
"You are a stupid, clumsy, few-headed piece of provender!"
Gop flushed clear down to his tails. "I'm sorry," he said. "I can't think
of everything at once! I must have accidentally activated the tharn
when I moved it. I'm sorry!"