International Business Machines Corporation

Naigang Wang

Eugene John O'Sullivan

Hariklia Deligianni

Bruce Bennett Doris

William Joseph Gallagher

Lubomyr Taras Romankiw

Bucknell Chapman Webb

Philipp Herget

Kailash Gopalakrishnan

Robert Fontana

Leland Chang

Rajiv V Joshi

Devendra K Sadana

Edmund Juris Sprogis

David Goren

Jungwook Choi

Chia-Yu Chen

Ankur Agrawal

Andrew John Kellock

Maurice Mason

Silvia Melitta Mueller

Guohan Hu

Xiao Sun

Chandrasekharan Kothandaraman

Todd Edward Takken

David Lee Rath

Xin Zhang

Daniel Brand

Joonah Yoon

Sathana Kitayaporn

Peilin Song

Xiao Lin Sun

Franco Stellari

Yang Liu

Jiamin Ni

An apparatus includes a floating-point gradient register; an integer register; a memory bank; and an array of processing units. Each of the units includes a plurality of binary shifters having an integer input configured to obtain corresponding bits of a 4-bit integer multiplicand, and a shift-specifying input configured to obtain corresponding bits in an exponent field of a 4-bit floating point multiplier. The multiplier is specified in a mantissaless four-bit floating point format including a sign bit, three exponent bits, and no mantissa bits. An adder tree has a plurality of inputs coupled to outputs of the plurality of shifters, and a rounder has an input coupled to an output of the adder tree. The integer inputs are connected to the integer register; the shift-specifying inputs are connected to the floating-point gradient register; and outputs of the rounders are coupled to the memory bank.

Four-bit training for machine learning

A micro-electromechanical device and method of manufacture are disclosed. A sacrificial layer is formed on a silicon substrate. A metal layer is formed on a top surface of the sacrificial layer. Soft magnetic material is electrolessly deposited on the metal layer to manufacture the micro-electromechanical device. The sacrificial layer is removed to produce a metal beam separated from the silicon substrate by a space.

Micro-electromechanical device having a soft magnetic material electrolessly deposited on a palladium layer coated metal beam

An apparatus includes circuitry for a neural network that is configured to perform forward propagation neural network operations on floating point numbers having a first n-bit floating point format. The first n-bit floating point format has a configuration consisting of a sign bit, m exponent bits and p mantissa bits where m is greater than p. The circuitry is further configured to perform backward propagation neural network operations on floating point numbers having a second n-bit floating point format that is different than the first n-bit floating point format. The second n-bit floating point format has a configuration consisting of a sign bit, q exponent bits and r mantissa bits where q is greater than m and r is less than p.

Neural network circuitry having floating point format with asymmetric range

An apparatus for training and inferencing a neural network includes circuitry that is configured to generate a first weight having a first format including a first number of bits based at least in part on a second weight having a second format including a second number of bits and a residual having a third format including a third number of bits. The second number of bits and the third number of bits are each less than the first number of bits. The circuitry is further configured to update the second weight based at least in part on the first weight and to update the residual based at least in part on the updated second weight and the first weight. The circuitry is further configured to update the first weight based at least in part on the updated second weight and the updated residual.

Machine learning hardware having reduced precision parameter components for efficient parameter update

Techniques regarding autonomous identification of aged circuits are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise an identification component, operatively coupled to the processor, that can identify an aged circuit by analyzing a current-voltage characteristic curve for a distortion in a sub-threshold quiescent current signature of the aged circuit.

Detection of an aged circuit

In an embodiment, a method includes configuring a specialized circuit for floating point computations using numbers represented by a hybrid format, wherein the hybrid format includes a first format and a second format. In the embodiment, the method includes operating the further configured specialized circuit to store an approximation of a numeric value in the first format during a forward pass for training a deep learning network. In the embodiment, the method includes operating the further configured specialized circuit to store an approximation of a second numeric value in the second format during a backward pass for training the deep learning network.

Hybrid floating point representation for deep learning acceleration

An apparatus includes a memory and a processor coupled to the memory. The processor includes first and second sets of arithmetic units having first and second precision for floating-point computations, the second precision being lower than the first precision. The processor is configured to obtain a machine learning model trained in the first precision, to utilize the second set of arithmetic units to perform inference on input data, to utilize the first set of arithmetic units to generate feedback for updating parameters of the second set of arithmetic units based on the inference performed on the input data by the second set of arithmetic units, to tune parameters of the second set of arithmetic units based at least in part on the feedback generated by the first set of arithmetic units, and to utilize the second set of arithmetic units with the tuned parameters to generate inference results.

Mixed precision capable hardware for tuning a machine learning model

Embodiments are directed to a method of forming a magnetic stack arrangement of a laminated magnetic inductor having a high frequency peak quality factor (Q). A first magnetic stack is formed having one or more magnetic layers alternating with one or more insulating layers in a first inner region of a laminated magnetic inductor. A second magnetic stack is formed opposite a surface of the first magnetic stack in an outer region of the laminated magnetic inductor. A third magnetic stack is formed opposite a surface of the second magnetic stack in a second inner region of the laminated magnetic inductor. The insulating layers are formed such that a thickness of an insulating layer in the second magnetic stack is greater than a thickness of an insulating layer in the first magnetic stack.

Laminated magnetic inductor stack with high frequency peak quality factor

Embodiments of the invention are directed to a method of fabricating a yoke arrangement of an inductor. A non-limiting example method includes forming a dielectric layer across from a major surface of a substrate. The method further includes configuring the dielectric layer such that it imparts a predetermined dielectric layer compressive stress on the substrate. A magnetic stack is formed on an opposite side of the dielectric layer from the substrate, wherein the magnetic stack includes one or more magnetic layers alternating with one or more insulating layers. The method further includes configuring the magnetic stack such that it imparts a predetermined magnetic stack tensile stress on the dielectric layer, wherein a net effect of the predetermined dielectric layer compressive stress and the predetermined magnetic stack tensile stress on the substrate is insufficient to cause a portion of the major surface of the substrate to be substantially non-planar.

Stress management for thick magnetic film inductors

Embodiments are directed to a method of forming a laminated magnetic inductor and resulting structures having multiple magnetic layer thicknesses. A first magnetic stack having one or more magnetic layers alternating with one or more insulating layers is formed in a first inner region of the laminated magnetic inductor. A second magnetic stack is formed opposite a major surface of the first magnetic stack in an outer region of the laminated magnetic inductor. A third magnetic stack is formed opposite a major surface of the second magnetic stack in a second inner region of the laminated magnetic inductor. The magnetic layers are formed such that a thickness of a magnetic layer in each of the first and third magnetic stacks is less than a thickness of a magnetic layer in the second magnetic stack.

Magnetic inductor with multiple magnetic layer thicknesses

A planar magnetic structure includes a closed loop structure having a plurality of core segments divided into at least two sets. A coil is formed about one or more core segments. A first antiferromagnetic layer is formed on a first set of core segments, and a second antiferromagnetic layer is formed on a second set of core segments. The first and second antiferromagnetic layers include different blocking temperatures and have an easy axis pinning a magnetic moment in two different directions, wherein when current flows through the coil, the magnetic moments rotate to form a closed magnetic loop in the closed loop structure.

Planar solenoid inductors with antiferromagnetic pinned cores

Embodiments are directed to a method of forming a laminated magnetic inductor and resulting structures having anisotropic magnetic layers. A first magnetic stack is formed having one or more magnetic layers alternating with one or more insulating layers. A trench is formed in the first magnetic stack oriented such that an axis of the trench is perpendicular to a hard axis of the magnetic inductor. The trench is filled with a dielectric material.

Magnetic inductor with shape anisotrophy

Method for forming a planar solenoid inductor

A method for fabricating a magnetic material stack on a substrate, comprises forming a first dielectric layer, forming a first magnetic material layer on the first dielectric layer, forming at least a second dielectric layer on the first magnetic material layer and forming at least a second magnetic material layer on the second dielectric layer. During one or more of the forming steps, a surface smoothing operation is performed to remove at least a portion of surface roughness on the layer being formed.

Growing community of inventors

Ossining, NY, United States of America

Naigang Wang