<s>
A	O
cognitive	B-General_Concept
computer	I-General_Concept
is	O
a	O
computer	O
that	O
hardwires	O
artificial	B-Application
intelligence	I-Application
and	O
machine-learning	O
algorithms	O
into	O
an	O
integrated	O
circuit	O
(	O
printed	O
circuit	O
board	O
)	O
that	O
closely	O
reproduces	O
the	O
behavior	O
of	O
the	O
human	O
brain	O
.	O
</s>
<s>
Synonyms	O
are	O
neuromorphic	B-General_Concept
chip	I-General_Concept
and	O
cognitive	B-General_Concept
chip	I-General_Concept
.	O
</s>
<s>
An	O
example	O
of	O
an	O
cognitive	B-General_Concept
computer	I-General_Concept
implemented	O
using	O
neural	B-Architecture
networks	I-Architecture
and	O
deep	B-Algorithm
learning	I-Algorithm
techniques	O
is	O
IBM	O
's	O
Watson	B-Application
machine	O
.	O
</s>
<s>
A	O
subsequent	O
development	O
by	O
IBM	O
is	O
the	O
2014	O
TrueNorth	O
microchip	O
architecture	O
,	O
which	O
is	O
designed	O
to	O
be	O
closer	O
in	O
structure	O
to	O
the	O
human	O
brain	O
than	O
the	O
von	B-Architecture
Neumann	I-Architecture
architecture	I-Architecture
used	O
in	O
conventional	O
computers	O
.	O
</s>
<s>
In	O
2017	O
Intel	O
also	O
announced	O
its	O
own	O
version	O
of	O
a	O
cognitive	B-General_Concept
chip	I-General_Concept
in	O
"	O
Loihi	O
"	O
,	O
which	O
it	O
intended	O
to	O
be	O
available	O
to	O
university	O
and	O
research	O
labs	O
in	O
2018	O
.	O
</s>
<s>
TrueNorth	O
was	O
a	O
neuromorphic	O
CMOS	B-Device
integrated	O
circuit	O
produced	O
by	O
IBM	O
in	O
2014	O
.	O
</s>
<s>
It	O
is	O
a	O
manycore	B-General_Concept
processor	I-General_Concept
network	B-Architecture
on	I-Architecture
a	I-Architecture
chip	I-Architecture
design	O
,	O
with	O
4096	O
cores	O
,	O
each	O
one	O
having	O
256	O
programmable	O
simulated	O
neurons	O
for	O
a	O
total	O
of	O
just	O
over	O
a	O
million	O
neurons	O
.	O
</s>
<s>
In	O
turn	O
,	O
each	O
neuron	O
has	O
256	O
programmable	O
"	O
synapses	B-Application
"	O
that	O
convey	O
the	O
signals	O
between	O
them	O
.	O
</s>
<s>
Hence	O
,	O
the	O
total	O
number	O
of	O
programmable	O
synapses	B-Application
is	O
just	O
over	O
268	O
million	O
(	O
228	O
)	O
.	O
</s>
<s>
Since	O
memory	O
,	O
computation	O
,	O
and	O
communication	O
are	O
handled	O
in	O
each	O
of	O
the	O
4096	O
neurosynaptic	O
cores	O
,	O
TrueNorth	O
circumvents	O
the	O
von	B-Architecture
Neumann-architecture	I-Architecture
bottleneck	O
and	O
is	O
very	O
energy-efficient	O
,	O
with	O
IBM	O
claiming	O
a	O
power	O
consumption	O
of	O
70	O
milliwatts	O
and	O
a	O
power	O
density	O
that	O
is	O
1/10	O
,	O
000th	O
of	O
conventional	O
microprocessors	O
.	O
</s>
<s>
The	O
SyNAPSE	B-Application
chip	O
operates	O
at	O
lower	O
temperatures	O
and	O
power	O
because	O
it	O
only	O
draws	O
power	O
necessary	O
for	O
computation	O
.	O
</s>
<s>
Skyrmions	O
have	O
been	O
proposed	O
as	O
models	O
of	O
the	O
synapse	B-Application
on	O
a	O
chip	O
.	O
</s>
<s>
The	O
said	O
cores	O
are	O
event-driven	O
by	O
using	O
both	O
(	O
a	O
)	O
synchronous	O
logic	O
and	O
are	O
interconnected	O
through	O
an	O
asynchronous	O
packet-switched	B-Protocol
mesh	O
network	B-Architecture
on	I-Architecture
chip	I-Architecture
(	O
NOC	B-Architecture
)	O
.	O
</s>
<s>
It	O
included	O
simulator	O
,	O
a	O
new	O
programming	O
language	O
,	O
an	O
integrated	B-Application
programming	I-Application
environment	I-Application
and	O
even	O
libraries	O
.	O
</s>
<s>
This	O
lack	O
of	O
backward	B-General_Concept
compatibility	I-General_Concept
with	O
any	O
previous	O
technology	O
(	O
e.g.	O
</s>
<s>
Intel	O
's	O
self-learning	O
neuromorphic	B-General_Concept
chip	I-General_Concept
,	O
named	O
Loihi	O
(	O
produced	O
in	O
2017	O
)	O
,	O
perhaps	O
named	O
after	O
the	O
Hawaiian	O
seamount	O
Lōʻihi	O
,	O
offers	O
substantial	O
power	O
efficiency	O
designed	O
after	O
the	O
human	O
brain	O
.	O
</s>
<s>
Intel	O
claims	O
Loihi	O
is	O
about	O
1000	O
times	O
more	O
energy	O
efficient	O
than	O
the	O
general-purpose	O
computing	O
power	O
needed	O
to	O
train	O
the	O
neural	B-Architecture
networks	I-Architecture
that	O
rival	O
Loihi	O
's	O
performance	O
.	O
</s>
<s>
In	O
theory	O
,	O
this	O
would	O
support	O
both	O
machine	O
learning	O
training	O
and	O
inference	O
on	O
the	O
same	O
silicon	O
independently	O
of	O
a	O
cloud	O
connection	O
,	O
and	O
more	O
efficient	O
than	O
using	O
convolutional	B-Architecture
neural	I-Architecture
networks	I-Architecture
(	O
CNNs	B-Architecture
)	O
or	O
deep	B-Algorithm
learning	I-Algorithm
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
Intel	O
points	O
to	O
a	O
system	O
for	O
monitoring	O
a	O
person	O
's	O
heartbeat	O
,	O
taking	O
readings	O
after	O
events	O
such	O
as	O
exercise	O
or	O
eating	O
,	O
and	O
using	O
the	O
cognitive	B-General_Concept
computing	I-General_Concept
chip	O
to	O
normalize	O
the	O
data	O
and	O
work	O
out	O
the	O
‘	O
normal’	O
heartbeat	O
.	O
</s>
<s>
The	O
first	O
iteration	O
of	O
the	O
Loihi	O
chip	O
was	O
made	O
using	O
Intel	O
's	O
14nm	O
fabrication	O
process	O
,	O
and	O
houses	O
128	O
clusters	O
of	O
1,024	O
artificial	B-Algorithm
neurons	I-Algorithm
each	O
for	O
a	O
total	O
of	O
131,072	O
simulated	O
neurons	O
.	O
</s>
<s>
This	O
offers	O
around	O
130	O
million	O
synapses	B-Application
,	O
which	O
is	O
still	O
a	O
rather	O
long	O
way	O
from	O
the	O
human	O
brain	O
's	O
800	O
trillion	O
synapses	B-Application
,	O
and	O
behind	O
IBM	O
's	O
TrueNorth	O
,	O
which	O
has	O
around	O
256	O
million	O
by	O
using	O
64	O
by	O
4,096	O
cores	O
.	O
</s>
<s>
Loihi	O
is	O
now	O
available	O
for	O
research	O
purposes	O
among	O
more	O
than	O
40	O
academic	O
research	O
groups	O
as	O
a	O
USB	B-Protocol
form	O
factor	O
.	O
</s>
<s>
Recent	O
developments	O
include	O
a	O
64	O
core	B-Architecture
chip	O
named	O
Pohoiki	O
Beach	O
(	O
after	O
Isaac	O
Hale	O
Beach	O
Park	O
,	O
also	O
known	O
as	O
Pohoiki	O
)	O
.	O
</s>
<s>
In	O
October	O
2019	O
,	O
researchers	O
from	O
Rutgers	O
University	O
published	O
a	O
research	O
paper	O
to	O
demonstrate	O
the	O
energy	O
efficiency	O
of	O
Intel	O
's	O
Loihi	O
in	O
solving	O
Simultaneous	B-Application
localization	I-Application
and	I-Application
mapping	I-Application
.	O
</s>
<s>
SpiNNaker	B-General_Concept
(	O
Spiking	B-Algorithm
Neural	I-Algorithm
Network	I-Algorithm
Architecture	O
)	O
is	O
a	O
massively	B-Operating_System
parallel	I-Operating_System
,	O
manycore	B-General_Concept
supercomputer	B-Operating_System
architecture	I-Operating_System
designed	O
by	O
the	O
Advanced	O
Processor	O
Technologies	O
Research	O
Group	O
at	O
the	O
Department	O
of	O
Computer	O
Science	O
,	O
University	O
of	O
Manchester	O
.	O
</s>
<s>
Critics	O
argue	O
that	O
a	O
room-sized	O
computer	O
–	O
like	O
the	O
case	O
of	O
Watson	B-Application
–	O
is	O
not	O
a	O
viable	O
alternative	O
to	O
a	O
three-pound	O
human	O
brain	O
.	O
</s>
<s>
In	O
2021	O
,	O
New	O
York	O
Times	O
released	O
the	O
article	O
"	O
What	O
Ever	O
Happened	O
to	O
IBM	O
’s	O
Watson	B-Application
?	O
"	O
.	O
</s>
<s>
They	O
wrote	O
about	O
some	O
costly	O
failures	O
of	O
IBM	B-Application
Watson	I-Application
.	O
</s>
<s>
During	O
the	O
collaboration	O
,	O
IBM	B-Application
Watson	I-Application
could	O
not	O
tap	O
patient	O
data	O
.	O
</s>
<s>
Watson	B-Application
struggled	O
to	O
decipher	O
doctors’	O
notes	O
and	O
patient	O
histories	O
.	O
</s>
