Story image

Toshiba develops a high-speed algorithm for deep learning processors

07 Nov 2018

Toshiba Memory Corporation announced the development of a high-speed and high-energy-efficiency algorithm and hardware architecture for deep learning processing with fewer degradations of recognition accuracy. 

The new processor for deep learning implemented on an FPGA achieves four times energy efficiency compared to conventional ones. 

Deep learning calculations generally require large amounts of multiply-accumulate (MAC) operations, and it has resulted in issues of long calculation time and large energy consumption. 

Although techniques reducing the number of bits to represent parameters (bit precision) have been proposed to reduce the total calculation amount, one of proposed algorithm reduces the bit precision down to one or two bit, those techniques cause degraded recognition accuracy. 

Toshiba Memory developed the new algorithm reducing MAC operations by optimising the bit precision of MAC operations for individual filters in each layer of a neural network. 

By using the new algorithm, the MAC operations can be reduced with less degradation of recognition accuracy.

Furthermore, Toshiba Memory developed a new hardware architecture, called the bit-parallel method, which is suitable for MAC operations with different bit precision. 

This method divides each various bit precision into a bit one by one and can execute 1-bit operation in numerous MAC units in parallel. 

It significantly improves the utilisation efficiency of the MAC units in the processor compared to conventional MAC architectures that execute in series.

Toshiba Memory implemented ResNet50, a deep neural network, on an FPGA using the various bit precision and bit-parallel MAC architecture. 

In the case of image recognition for the image dataset of ImageNet, the above technique supports that both operation time and energy consumption for recognising image data are reduced to 25 % with less recognition accuracy degradation, compared to conventional method.

Artificial intelligence (AI) is forecasted to be implemented in various devices. The developed high-speed and low-energy-consumption techniques for deep-learning processors are expected to be utilised for various edge devices like smartphones and HMDs and data centres which require low energy consumption. 

High-performance processors like GPU are important devices for high-speed operation of AI. 

Memories and storages are also one of the most important devices for AI which inevitably use big data. 

Toshiba Memory Corporation is continuously focusing on research and development of AI technologies as well as innovating memories and storages to lead data-oriented computing.

Protecting data centres from fire – your options
Chubb's Pierre Thorne discusses the countless potential implications of a data centre outage, and how to avoid them.
Opinion: How SD-WAN changes the game for 5G networks
5G/SD-WAN mobile edge computing and network slicing will enable and drive innovative NFV services, according to Kelly Ahuja, CEO, Versa Networks
TYAN unveils new inference-optimised GPU platforms with NVIDIA T4 accelerators
“TYAN servers with NVIDIA T4 GPUs are designed to excel at all accelerated workloads, including machine learning, deep learning, and virtual desktops.”
AMD delivers data center grunt for Google's new game streaming platform
'By combining our gaming DNA and data center technology leadership with a long-standing commitment to open platforms, AMD provides unique technologies and expertise to enable world-class cloud gaming experiences."
Inspur announces AI edge computing server with NVIDIA GPUs
“The dynamic nature and rapid expansion of AI workloads require an adaptive and optimised set of hardware, software and services for developers to utilise as they build their own solutions."
365 Data Centers secures additional funding for expansion
The company asserts the financing commitments come as it looks to invest in further substantial internal and external growth.
HPE launches 'right mix' hybrid cloud assessment tool
HPE has launched an ‘industry-first assessment software’ to help businesses work out the right mix of hybrid cloud for their needs.
ADLINK and Charles announce multi-access pole-mounted edge AI solution
The new solution is a compact low profile pole or wall mountable unit based on an integration of ADLINK’s latest AI Edge Server MECS-7210 and Charles’ SC102 Micro Edge Enclosure.