HAL Id: jpa-00245508
https://hal.archives-ouvertes.fr/jpa-00245508
Submitted on 1 Jan 1987
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Power supply constraints in megabit DRAMs of the future
K. Itoh, K. Kimura
To cite this version:
K. Itoh, K. Kimura. Power supply constraints in megabit DRAMs of the future. Re- vue de Physique Appliquée, Société française de physique / EDP, 1987, 22 (1), pp.15-19.
�10.1051/rphysap:0198700220101500�. �jpa-00245508�
Power supply constraints in megabit DRAMs of the future
K. Itoh and K. Kimura
Central Research
Lab.,
Hitachi Ltd,Kokubunji, Tokyo
185,Japan
(Reçu
le 28juin
1985,accepté
le 9juin 1986)
Résumé. 2014 La densité des mémoires DRAM a été
multipliée
par quatre tous les trois ans. Ledéveloppement
desmémoires 1 Mb est d’ores et
déjà
très avancé. L’évolution de l’alimentation est doncd’importance capitale
étantdonné les réductions des dimensions
qui
sont intervenues.Abstract. 2014
Megabit
DRAM powersupply
is described in terms of powerdissipation, reliability
for small transistors, and memory celloperating margin.
Suchrecently developed
1 Mbtechniques
as CMOS andvertically
structuredmemory cells are discussed. It is concluded that, in
spite
of CMOSadvantage,
a transition from theexisting supply voltage
of 5 Vmight
occur at the 16 Mb level.Classification
Physics
Abstracts85.40
1. Introduction.
The
density
ofdynamic
random-access memories(DRAMs)
hasquadrupled
every three years. In line withthis,
thedevelopment
of 1 Mb DRAMs[1-12]
asthe next
step
from theexisting
256 kb isalready
underway. This
rapid
evolution in DRAMs has been realizedmainly by
the progress inphotolithography
and one-device cell. Such progress has made it clear that one of the
major
concems in themegabit
DRAMsdesign
inthe near
future
is the issue of powersupply voltage.
This issue is considered in this paper.
First,
it is shown how thesupply voltage
isclosely
related to powerdissipation
and submicrometer transistor character-istics,
which constrain the upper limit of thesupply voltage. Next,
a one-device celloperating margin,
which constrains the lower limit of the
supply voltage,
is
discussed,
followedby
future considerations concem-ing
the powersupply.
2. Power
dissipation.
There is a definite trend of increased power
dissipation
with
increasing
bitdensity
for agiven supply voltage, V cc’
as shown infigures
1 and 2. This is due toincreases in both
peripheral
current,ip,
and in memoryarray current,
i A,
as shown infigures
3 and 4. It shouldbe noted that the contribution of
i A
to the total powerdissipation
becomes more and morepronounced
as thebit
density increases.
The increase inip
is causedby
increased load
capacitance
inperipheral
circuits due toan increase in the number of decoders and an increase in
chip
size of about 1.5 times for every successiveREVUE DE PHYSIQUE APPLIQUÉE.-T. 22, N’ 1, JANVIER 1987
Fig.
1. - Trends insupply voltage
and powerdissipation.
Fig.
2. - Powerdissipation analysis
for NMOS DRAMs.2
Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/rphysap:0198700220101500
16
Fig.
3. - Factorsconstraining
thesupply voltage
in MbDRAMs.
Fig.
4. -Logical
DRAM memory arrayconfiguration.
generation
DRAM[13].
The increase ini A
is due to theincrease in total data line
capacitance, mnCD.
In orderto offset these
capacitance increases,
power reductiontechniques
have beenemployed
in eachgeneration
DRAM : 16
K ; j8
was reducedby
half due to adoubling
of n. Inaddition, dynamic
senseamplifiers [14]
wasintroduced,
64K ;
powerdissipation [15]
wasdrastically
reducedcompared
to theprevious
gener-ation, by reducing
thesupply voltage
from 12 V to 5 Vwhile
maintaining
the sametREF max/n,
256K ; tREF
max was doubled while
j8
remained the same[16].
However,
this has not been the case with the 1 Mbgeneration. Reducing
thesupply voltage
to 3.3 V wasproposed
in theearly stage
of 1 Mbdevelopment
toreduce power
dissipation
and to insurehigh reliability
in small transistors.
Eventually,
aVcc
of 5 V waschosen as a result of the
following
low powertechniques
combined with a further increase in
tREF
max : the useof CMOS
[3-7]
instead of the traditionalNMOS,
theuse of a half
Vcc precharge
on the data lines[2, 4-9],
the use of an
on-chip voltage
limiter circuit[1, 3]
andthe use of multidivision data lines
[1, 4, 9].
It was not until the 1 Mb
generation
of DRAMs that CMOS DRAMs[17-19]
wereseriously
taken into consideration. This isbecause,
aslong
as NMOS isused,
it is too difficult todesign
a low power 1 Mbchip
in
spite
of a refreshcycle,
n, which has increased from256 to
512,
as shown infigure
2. The use of CMOSclock drivers and decoders in
peripheral
circuits dramat-ically
reducesi p by
about half.Furthermore,
the use of CMOSpermits
theimplementation
of low powerspecial
functions such as a static column mode. The data lineprecharged voltage, Vp,
can be reducedby
half
using
a halfVcc precharge. This,
in turn, results ina reduction of the array current,
i A,
as shown in theequation
infigure
3. Thisapproach
is desirable forapplication
in CMOS because it facilitates an area-saving layout
as well as offers a wideoperating margin
for
amplifiers
and active restore circuits. Another way to reduceVp
has also beenproposed :
memory arrayoperation
should be based on a reducedvoltage, generated by
anon-chip
limiter while thevoltage supplied externally
is maintained at 5 V.ip
can also bereduced very
effectively by reducing CD.
This can beaccomplished by using
a multi-divided data line struc- ture. This reduction inCD
also results in an increased memory cellsignal voltage.
Such a data line structure,using
two-level aluminumwiring,
has beenreported
toreduce
by
2/3 the data linecapacitance
for agiven chip
area.
Despite
thesepast attempts
to achieve low powerdissipation
whilemaintaining
a 5 Vsupply,
it willbecome more and more difficult to reduce the power
dissipation
without furtherreducing
theVcc.
3. Submicrometer transistor characteristics.
As device dimensions continue to shrink to the submic-
ron
level,
concerns arise that such devices will nolonger
be able to withstand the enhanced electric fieldsimposed by
a 5 V powersupply.
Infact,
hot-carrierinjection
and thepossibility
ofperformance degradation
associated with it have been
reported
even for DRAMas small as 64 kb
[20].
As aresult,
stress-immune device structures[21]
such asAs-P
double diffused drain(DDD)
andlightly doped
drain(LDD)
have beenintensively
studied andapplied
to actual 1 Mbchips [5].
Even if future
developmental
efforts are focused on thepursuit
of other new submicrondevices,
it ispossible
toassume that it will be inevitable to reduce the
Vcc,
eitherinternally
orexternally,
from theexisting
5 V. This is
mainly
because it will be difficult to fabricate new stress-immune submicrometer devices which can withstand suchhigh supply voltage (5 V)
without
employing
verycomplicated
andcostly
proces-ses. In
addition,
the relative conductance of such devices will behigher
at a lowersupply voltage [22].
4.
Memory
cellmargin.
In the
light
of theprevious arguments,
it seemsappropriate
to reduce thesupply voltage
formegabit
DRAMs.
However,
thesupply voltage
isstrictly
limitedby
the memory cellmargin.
The cellmargin
can beroughly expressed
as the sum of thenoise, VN,
causedby
the electrical unbalance in the data lineitself,
thecapacitive coupling
from theadjacent
data lines and a senseamplifier
offsetvoltage,
and the storedvoltage degradations, VL
andV a,
due toleakage
current and aparticle, respectively,
as shown infigures
3 and 5.Fig.
5. - Schematic SINdesign
for DRAM cell.V p
=V,/2:
halfVp precharge, V 0
=V cc :
no limitercircuit.
In order to assure a sufficient cell
margin,
ahigher supply voltage
isobviously
better.Thus,
a furtherreduction in
supply voltage
necessitates the use of innovative cell structures which allow lowerVN, VL
and
V a.
To meet thisrequirement,
a tremendous effort isbeing
made todevelop
advanced memory cells[23, 24]
which exceed theperformance
ofcommercially accepted
memory cells.Among
these newcells,
verti-cally
structured cells[1, 3, 4, 8-10, 25, 26] having high Cs
such as trench cells and stackedcapacitor cells,
arebecoming increasingly important.
Inaddition,
aplanar
cell structure combined with thin
Sio2
filmoperating
athalf
VCC plate voltage [5, 6, 8,10,11]
is alsoextremely
efficient to obtain
high Cs.
A folded data line cellincorporating
a multidivision data line and a halfycc precharge provides
the lowestVN.
Anepitaxial layer [27]
and a stackedcapacitor
cell arereported
tobe effective in
reducing leakage
current. Note that thereduction of power
dissipation
is also essential forreducing
theleakage
current.Many attempts
have been made to increase soft errorimmunity,
thatis,
to reducethe
critical, charge, Qc.
Theseattempts
include the useof such structures as Hi-C
[28],
anepitaxial layer [29],
aburied
layer
structure[30]
and stackedcapacitor [25].
The
p-channel
MOSswitching
transistor cell[17]
em-bedded in an n-well has also been
proposed
as anothermeans of
reducing
the soft error. In any event, the mainemphasis
must beput
on the memory cell in order to reduce thesupply voltage.
5.
Perspective
of powersupply voltage.
Continued progress on the road toward
megabit
DRAMs will
surely require
a reduction inchip Vcc.
How far thenbeyong
the 1 Mb DRAMs will a 5 Vsupply
survive ? This answer ishighly controversial,
ascan be
anticipated
from theprevious
discussion. This isbecause,
in addition to alternatives describedabove,
there are incalculable unforeseen
developments
in thefuture. What makes the matter more
complicated
is thefact that device feature size tends to be
slightly larger
than one would have
expected.
This is because thechip
size can be seen to increase with every
generation.
Dataobtained from actual
experimental
NMOS[1}
andCMOS
[4]
1 Mbchips (Table I)
seem to be useful inpredicting
the nextVcc
level. Bothchips
weredesigned using
almost the samelayout
rules to cover sometechniques
which arethought
to be useful in the future for low power,high
transistorreliability
and a wideoperating margin
for memory cells.Anticipated
powerdissipations
formegabit DRAMs,
based on thedesign
data from
chips
in tableI,
are shown with dashed lines infigure
1. Thefollowing assumptions
are made : theuse of such circuit
techniques
as a halfVcc precharge
and a multidivision data line as shown in table
1 ;
increase inchip
size of 1.5 times for every successivegeneration ;
the use of thesimply
scaled-down devices and memory cellshaving
a sufficient break downvoltage
and a wideoperating margin, respectively.
Itcan be seen that CMOS offers the best solution.
However,
atpresent,
even CMOS will not be able tohandle the
higher
powerdissipation
inmegabit
DRAMs unless the
supply voltage
is reduced and/or the number of refreshcycle
is increased. A criticalpoint
forthe power
supply
transition isthought
to be at the16 Mb level as far as power
dissipation
is concemed.Note that the
development
of both submicrometer transistors and memory cells to meet therequirement
for such
high density
DRAMs will become more andmore
important. Hence,
a 5 V powersupply might
survive even for the 4 Mb
generation through
thesimultaneous advancement of process, device and cir- cuit
development
which are still notfully developed
inthe 1 Mb
generation.
6. Conclusion.
It was shown that the
supply voltage
isclosely
related topower
dissipation, reliability
and the memory celloperating margin. Through discussion,
it was concludedthat,
inspite
ofCMOS,
a transition from theexisting
supply voltage
of 5 Vmight
occur at the 16 Mb level.18
Table I. -
Performance comparison of
1 Mb NMOS and CMOS DRAMS.References
[1]
ITOH, K. etal.,
Anexperimental
1 Mb DRAM withon-chip voltage
limiter, IEEE Int. Solid-State Circuits Conf. Tech.Dig.,
p. 282, Feb., 1984.[2] SUZUKI,
S. et al., A 128 K word 8 bitdynamic
RAM, ibid.,p. 106.
[3]
YAMADA, J. etal.,
A submicron VLSI memory with a bit-at-time built- in ECC circuits,ibid., p. 104.
[4]
SATO, K. etal.,
A 20 ns static column 1 Mb DRAM in CMOStechnology,
IEEE Int. Solid-State Cir- cuits Conf. Tech., p. 254, Feb., 1985.[5] SAITO,
S. etal., A 1
Mb CMOS DRAM withfast
page and static columnmodes, ibid.,
p. 252.[6] TAYLOR,
R. and JOHNSON, M., A 1 Mb CMOS DRAM with a divided bit line matrixarchitecture, ibid.,
p. 242.[7] KIRSCH,
H. C. etal. ,
A 1 Mb CMOS DRAM,ibid., p. 256.
[8]
TAKEMAE, Y. etal. ,
A 1 Mb DRAM with 3-dimen- sional stackedcapacitor cells, ibid.,
p. 250.[9]
INOUE, Y. etal.,
An 85 ns 1 Mb DRAM in aplastic DIP, ibid.,
p. 238.[10]
HORIGUCHI, F. et al., A 1 Mb DRAM with afolded capacitor
cell structure,ibid.,
p. 244.[11]
KUMANOYA, M. etal.,
A 90 ns 1 Mb DRAM withmulti-bit test
mode, ibid.,
p. 240.[12]
KALTER, H. L. etal.,
Anexperimental
80 ns 1 MbDRAM with
fast
pageoperation, ibid.,
p. 248.[13]
NOBLE, W. P. and WALKER,W. W.,
Fundamental limitations on DRAM storagecapacitors,
IEEECircuits Devices
Mag. (1985)
45.[14] SCHROEDER,
P. R., and PROEBSTING, R. J., A 16 K x 1 bitdynamic
RAM, IEEE Int. Solid- State Circuits Conf. Tech.Dig., p. 12,
Feb., 1977.[15]
ITOH, K. etal. ,
Asingle
5 V 64 Kdynamic
RAM,IEEE Int. Solid-state Circuits Conf. Tech.
Dig.,
p. 228, Feb., 1980.
[16]
MATSUE, S. etal.,
A 256 Kdynamic
RAM,ibid., p. 232.
[17] SHIMOHIGASHI,
K. etal.,
An n-well CMOSdynamic
RAM, IEEE Electron Devices Meet. Tech.Dig.,
p. 835, Dec. 1980.[18] CHWANG,
R. etal.,
A 70 nshigh density
CMOS DRAM, IEEE Int. Solid-State Circuits Conf.Tech.
Dig.,
p. 56, Feb., 1983.[19] KAWAMOTO,
H. etal.,
A 288 Kb CMOSpseudo SRAM,
IEEE Int. Solid-State Circuits Conf.Tech.
Dig.,
p. 276, Feb., 1984.[20] YAMADA,
M. etal. ,
Hot-electrontrapping effects of
short channel 64 k
dynamic
MOSRAM,
Proc.14th Conf.
(1982 International)
Solid State De-vices, p. 17-18, Aug.,
1982.[21] TAKEDA,
E., Hot-carrier effects in submicrometer MOS VLSIs, IEE Proc.131,
Pt. I, No. 5(1984)
153-162.
[22] TAKEDA,
E., Constraints on theapplication
on0.5 03BCm MOSFET’s to ULSI systems, IEEE J.
Solid-State Circuits SC-20, No. 1
(1985)
242-247.[23]
ITOH, K. and SUNAMI, H.,High-density
one-devicedynamic
MOS memory cells, IEE Proc.130,
Pt. I, No. 3
(1983)
127-135.[24]
ISHIHARA, M. and KAWAMOTO, H., Trends inhigh density
DRAMs, ESSCIRC 84,Digest
of Techni-cal
Papers.
[25]
KOYANAGI, M. et al., A 5 Vonly
16 K bit stacked-capacitor
MOS RAM, IEEE J. Solid-State Cir- cuitsSC-15,
No. 4(1980)
661.[26]
SUNAMI, H. etal.,
Acorrugated capacitor
cell(CCC),
IEEE Trans. Elect. Devices, ED-31, No. 6(1984)
746.[27]
SATOH, S. et al. ,High capacitance
trench structure(HI-CAT)
formegabit
LSI, 1984Symposium
onVLSI
technology
Tech.Dig., p. 18, Sept., 1984.
[28]
YAMADA, M. etal., Soft
errorimprovement of dynamic
RAM with Hi-C structure, IEEE Elec- tron Devices Meet. Tech.Dig.,
p.578,
1980.[29] SATOH,
S. etal., Soft
errorimprovement
in MOSRAMs
by
the useof epitaxial substrate,
Proc.12th Conf. Solid State Devices,
p. 143, Aug.,
1980.
[30] DENNARD,
R. H., etal., Modeling
and controlof alpha-particle effects
in scaled-down VLSI cir- cuits, 1981symposium
on VLSItechnology,
Tech.