Page 22 - European Energy Innovation - Spring 2015 publication
P. 22
Spring 2015 European Energy Innovation


Reducing the “Big data” ogre’s energy

New applications use of medium/low-throughput and will promote a “global
in the Internet of network such as the one recently system approach” whereupon
Things (IoT), which proposed by SigFox. STT-MRAM technology is also
promises to change embedded within the logic blocks
our lives, will require storing As node-level individual devices themselves. Together with specific
and computing ever-increasing are often distributed in remote firmware libraries, this will result in
amounts of data. By 2020, it is and/or difficult to access area, a massive decrease of the overall
expected that IoT will generate power is seldom available and chip power consumption both in
around 4.4 zettabyte of data batteries life is crucial. Moreover, active (operating) and stand-by
worldwide (source: IDC 2014). To these systems are mostly idle modes.
transfer, sort and analyze those (90 to 99.9% of the time), hence
“Big Data”, it will cost, per year efficient on-chip storage implies Thanks to this new approach,
and only in the US and only for the use of non-volatile memories eVaderis products will combine
the infrastructure (network and (e.g. memories which retain high performances and
data centers), 14 billion kilowatt data when power is turned off). intelligence levels, together with
hours, which accounts for $1.3 However a proper balance has to long lifetime operation, whilst also
billion and 15 million tons of CO2. be found between computing/ offering a highly flexible platform
storage intelligence and battery to software developers through,
Addressing this issue implies a life, as the more on-chip data, the for instance, quasi instant-on
drastic reduction of the amount higher the power consumption capabilities, transparent data/
of data that has to flow through and in return the lower the overall code partitioning and over-the-
the network, which translates into battery life. air ultra-low power hardware
the ability to perform distributed programmability. A smartmeter
data storage and processing at The solution proposed by embedding such a technology
the individual node/connected young start-up eVaderis, spin- would, for example, exhibit
device. As such, each node off of SPINTEC-lab, consists in high-processing and storage
now acts as an intelligent buffer integrating emerging non-volatile capabilities with 10 times less
between the sensing device and memory technologies such as standby power consumption and
the network, in turn enabling the STT-MRAM into microcontrollers/ 10.000 times less storage energy,
microprocessors used in wireless, over the air firmware update and
battery-operated devices. STT- enhanced security.
MRAM, by its sheer magnetic
nature, is intrinsically non volatile, All this of course has a price:
which existing Static-RAM Beyond the processing add-
(SRAM) memory is not, with a cost, it requires a total rebuilt
much lower active power than of the circuit and system level
FLASH, today’s best-in-class architectures. The pain is severe
non-volatile memory. It is this but the reward is there with a
unique combination which allows massive reduction of the IoT
to embed large amounts of ecosystem overall data path and
(non-volatile) memory at minimal energy footprint, from the devices
power cost. eVaderis will go to the network (transmission
beyond the simple SRAM/FLASH channels and server farms). The
memory blocks replacement future is here ! l
   17   18   19   20   21   22   23   24   25   26   27