 |
Description  |
|
|
FIELD OF THE INVENTION
This invention relates to packaging configurations for integrated circuit
devices (ICs) and particularly to an improvement to the design of a memory
array which requires fewer random access memories to be turned on during a
read or write cycle than present designs, thereby using less current.
BACKGROUND OF THE INVENTION
Current generation single in-line memory modules (SIMMs) for certain brands
of computers use eight one-megabit (1M) dynamic random access memories
(DRAMs) arranged in a .times.1 configuration (having one data out signal),
which supplies the computer with one megabyte (MB) of memory. Since the
DRAMs are arranged in a .times.1 configuration, one data bit can be
extracted from each chip at a time. When a module with eight 1M.times.1
DRAMs is installed in a computer capable of handling eight bits of data at
a time (i.e. an 8-bit computer), it accesses one bit location from each of
eight DRAMs on a module simultaneously, thereby receiving the eight bits
of data it is capable of handling. In 16-bit computers, modules containing
eight 1M.times.1 DRAMs are installed in groups of two in the computer. To
obtain the 16 bits of data the computer is capable of handling, all 16
DRAMs are accessed simultaneously, and the computer receives one bit of
data from each DRAM for a total of 16 data bits. Each time a 1M.times.1
DRAM is accessed, it requires about 80 mA of current to be supplied. To
access the 16 DRAMs simultaneously requires approximately 640 mA of
current per module, or 1,280 mA total.
Some SIMMs use 1M.times.4 DRAMs, with each DRAM having four bits of data. A
module using two 1M.times.4 chips supplies 1 mB of memory, as does a
module using eight 1M.times.1 chips. A module with two 1M.times.4 devices
is functionally equivalent to a module using eight 1M.times.1 devices, but
has fewer parts, thereby being easier to assemble and somewhat more
reliable due to fewer solder joints. There is not much power savings using
a module with two 1M.times.4 DRAMs over a module using eight 1M.times.1
DRAMs, as all the devices on either module are turned on each time one of
the devices is accessed in order to access eight data bits, and to access
two 1M.times.4 DRAMs requires about as much power as accessing eight
1M.times.1 DRAMs.
In most computers, addressed words are an even number of bits, such as
eight, sixteen or thirty-two bits. This fits neatly into memory array
blocks which use .times.4 chips. This convenient arrangement is
complicated by the fact that a system of memory parity has proven to be
very effective in error detection. The parity is an additional bit for
each word, so that an eight bit word ("byte") is addressed as nine bits,
the ninth bit being parity.
Reducing power consumption in a computer or other electronic device is a
design goal, as overtaxing a computer's power supply is a common concern.
With the addition of modem cards, memory boards, graphics cards, hard disk
controller cards, printer buffer cards, and mouse cards, the chances of
burning out the computer's power supply from drawing too much current
becomes a possibility. Even if the power supply is not unduly stressed, a
component which uses more power than a similar component will release more
heat, thereby increasing the temperature of the component as well as the
inside of the computer or electronic device. Elevated temperatures within
the component or within the chassis of a computer can cause other
components in the computer to operate more slowly or to fail prematurely.
Reducing the amount of current used by the components in a computer is also
a concern to designers of portable computers. The length of time between
battery recharges for various brands and types of computers ranges from
about two hours to 12 hours. Reducing the amount of current the computer
uses, thereby extending the length of time the computer can be run off the
battery, is a design concern as well as a marketing concern.
For the reasons listed above, reducing the power consumption of components
installed in a computer is a goal of computer component designers and
computer manufacturers.
SUMMARY OF THE INVENTION
An object of this invention is to provide a memory array which uses less
power than previous arrays.
This object of the present invention is attained by fabricating an array
using a number of memory chips, where each memory chip can be accessed
independently, and where for example only the DRAM or DRAMs accessed is
turned on while all other DRAMs remain in standby mode. A DRAM in standby
mode uses much less current than activating the DRAM.
The invention can be applied to modules using DRAMs with multiple data out
lines (DQ's) For instance, if a module supplying 1 mB of memory contains
eight 1M.times.1 DRAMs is installed in an 8-bit computer, all eight DRAMs
would have to be accessed simultaneously to supply the computer with 8
bits of data. On a 1 mB module using eight 256K.times.4, only two DRAMs
would have to be accessed to supply the 8-bit computer with 8 bits of
data.
Chips containing .times.16 data widths have recently been developed by
Micron Technology, Inc. To manufacture these 64K.times.16 DRAMs, a current
generation 1M die is packaged with 16 DQ pins to provide a chip in a
64K.times.16 configuration. Each of the 1,048,576 bits are uniquely
addressed through the 16 address bits multiplexed on eight address lines
(A0-A7) during a read or a write cycle.
A common memory configuration supplying 16 bits of data is to use two
modules with each module comprising eight 1M.times.1 devices. A read cycle
from two of these modules, as stated previously, requires about 640 mA of
current. A functional equivalent of these modules would be two modules
with each module comprising eight 64K.times.16 DRAMs. If these equivalent
modules not comprising the invention are used, all 16 DRAMs would be
turned on during a read cycle, even though the desired data comes from a
single DRAM. A read would require 1280 mA of current. A module of this
type comprising the invention, however, would enable only one DRAM during
a read, thereby using about 90 mA of current.
When used in applications where an additional bit is used, as for parity,
the additional bit may either be incorporated into the multiple data out
(DQ) architecture as an additional DQ connection. Alternatively, partially
operational DRAMs may be used, provided at least one good sector may be
addressed.
A module of this type would have signals conforming to JEDEC standards or,
in custom uses, to specifications specific to the intended use of the
module. In any case, a module containing eight 64K.times.16 devices would
require one CAS line and eight RAS lines. The CAS line selects the desired
column number in each of the eight DRAMs. The RAS lines are used as a bank
select with each RAS line being used only by a single device, thereby
accessing a row address from a single DRAM.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 shows an overview of the circuitry of the inventive module;
FIG. 2 details the decode circuitry of FIG. 1;
FIG. 3 shows a simple circuit which disables the write-per-bit mode of a
DRAM containing multiple DQ's;
FIG. 4 shows the logic associated with the signals AR9 and AC9 which
selects one of four groups of RAS signals;
FIG. 5 shows the logic associated with the signals AR8 and AC8 which
selects a single DRAM from a group of four DRAMs; and
FIG. 6 shows the logic associated with the write-per-bit lockout circuit of
FIG. 3.
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 shows an overview of the circuitry of one embodiment of the
inventive array, including the data bus (DQ1-DQ16), the address bus
(AD0-AD7), and output enable (OE). The address bus allows the computer to
select individual DRAM cells to be written to or read from, while data is
passed between the computer and the DRAMs along the bi-directional data
bus. The OE signal controls the output buffers of the DRAM. During a READ
cycle, the data is output on the data bus when the OE signal goes low.
All DRAMs share a single write enable (WE) signal, a single V.sub.cc, a
common V.sub.ss, and a common CAS.
The signals AR8, AR9, AC8, and AC9 output by the computer to the module are
altered by the decode circuitry (described below) to function as 16 RAS
lines, which select one of 16 DRAMs on the module. Note that FIG. 1 shows
eight RAMs; RAMs 9-16 operate in a fashion similar to RAMs 1-8, being
addressed by RAS9-RAS16 as shown.
Following JEDEC standards, a computer or electronic device has only one RAS
and one CAS input to a memory module. With only these two inputs, every
time the electronic device accesses the memory, the same address on every
DRAM is read or written, and as a result every device turns on. In a
module comprising .times.1 DRAMs, this is not a problem because, as stated
previously, a 16-bit computer needs to access all 16.times.1 DRAMs in
order to receive the 16 bits of data it is capable of handling. In a
module comprising DRAMs with multiple DQ's, however, not every DRAM is
accessed, but every DRAM is turned on. This requires that power be used
unnecessarily.
In the inventive module, turning on all the DRAMs would defeat the purpose
of the invention, which is to save power by turning on only those RAMs
that are accessed. The decode circuitry in FIG. 2 solves this problem by
using the two RAS address select bits (AR8 and AR9) and the two CAS
address select bits (AC8 and AC9) output from the computer to the module
in conjunction with the decode circuitry of FIG. 2 to turn on a single
device. AS shown, the two bits input on AR9 and AC9 are used to select one
of four RAS signals internal to the decode circuitry, RASA, RASB, RASC, or
RASD, depending on the state of the two bits as shown in FIG. 4.
Each of the four groups of signals in FIG. 4, RASA, RASB, RASC, and RASD
have four unique RAS signals as shown in FIG. 2 which are internal to the
decode circuitry and are output to the DRAMs. Referring to FIG. 2, after
either RASA, RASB, RASC, or RASD is turned on, the bits supplied on AR8
and AC8 are used to select a single location from RAS1 through RAS16, each
RAS line corresponding to a unique DRAM (not shown). FIG. 5 shows the
decode logic which selects a specific DRAM. As shown in FIG. 2, RASA is
divided into RAS1-RAS4, RASB is divided into RAS5-RAS8, RASC is divided
into RAS9-RAS12, and RASD is divided into RAS13-RAS16. So, for example, if
AR9 goes high and AC9 is a low, the signal RASC goes high. Then, if both
AR8 and AC8 go high, RAS12 goes high and accesses its associated DRAM,
thereby leaving RAS1-RAS11 and RAS13-RAS16 unselected and the 15 DRAMs
corresponding to those RAS lines in a power-conserving standby mode.
Write-per-bit mode is an industry standard on DRAMs having multiple DQ's. A
DRAM with multiple DQ's can be written to in either a normal write mode or
in write-per-bit mode. When a DRAM with more than one DQ is in a normal
write mode, the number of bits corresponding to the number of DQ's are
written at the same time. On a .times.16 device, (a device having 16
subarrays) for example, the chip logic begins writing one bit of data onto
each of the 16 DQ's at the falling edge of CAS or WE (whichever is later)
as long as RAS is low. (During a normal write, the status of WE is a
"don't care" when RAS initially goes low.) The address signals, RAS, and
CAS then toggle to select the proper address to be written to, and the
desired data is input through the Data In (Din) signals.
During a write-per-bit (also called a "masked write"), any combination (or
even all) of the 16 bits can be written to without writing to any of the
other locations. To set up a write-per-bit signal, WE goes low. Next, the
data for the "mask" is set on the DQ's, with a logic 1 corresponding to
"write" and a logic 0 corresponding to a "don't write" (the mask data
simply indicates which of the locations are to be written, and which are
to be left unaltered). After the data for the mask is set, RAS drops, and
the mask information on the data lines is changed to the desired data to
be written to the selected locations. Finally, when CAS is pulled low, the
write begins. The address signals, RAS, and CAS toggle to input the data
into the correct addresses.
As can be seen from the above, users of memory modules which
contain.times.1 DRAMS which don't use write-per-bit mode may consider WE a
"don't care" as RAS goes low, and allow WE to toggle. Depending on the
state of the other signals, the unwary user may put the module containing
DRAMs with multiple DQ's into write-per-bit mode (which, as previously
stated, occurs at the DRAM level if RAS goes low when WE is low). The
simple circuit of FIG. 3, if incorporated into the decode circuitry of the
module or into the design of the electronic device using the inventive
module, will make the WE signal a don't care except when RAS is low,
thereby preventing the chips on the modules from entering write-per-bit
mode. The circuit incorporates a three input NAND gate 10. RAS, WE, and a
RAS signal delayed by the three NAND gates 12, 14, 16 as shown in FIG. 3
are inverted, input to the NAND gate 10, and output as WE(out). (Note that
three NAND gates is not an absolute--the number of NAND gates is
determined only by the delay required to ensure that WE does not go low
until after RAS goes low.) The truth table for the circuit of FIG. 3 is
shown in FIG. 6.
A jumper, electronic switch, or a functional equivalent 18 incorporated
into the circuit would allow users who desire the write-per-bit mode to
disable the circuit, thereby enabling write-per-bit mode to the DRAMs.
While a preferred embodiment of the invention has been disclosed, various
modes of carrying out the principles described herein are contemplated as
being within the scope of the following claims. Any memory array
comprising RAMs (SRAMs, DRAMs, etc.) having multiple DQ's could have a
power savings by using the invention. For instance, in 1 mB module
comprising eight 256K.times.4 RAMs, all eight DRAMs are turned on for each
read, even though the 16 bits of data are received from only four of the
DRAMs. The description of the invention could be easily modified by those
skilled in the art for a .times.4 module.
In addition, modules with data widths other than those which are a multiple
of four are possible with the addition of another device, such as
a.times.1 device. For example, a .times.17 module is possible on a module
containing 64K.times.16 devices with the addition of a 64K.times.1 device.
Note that this device would require another RAS line, but would use the
common CAS signal, and at least two devices would be turned on
simultaneously to access the 17 bits of data required, one .times.16 DRAM
for the 16 data bits, and the .times.1 device for the parity bit.
Finally, the described invention doesn't pertain only to memory supplied in
module form. The invention would work equally well with memory placed
directly on the motherboard (embedded memory) or with any other memory
addressed by the computer.
It is therefore understood that the scope of the invention is not to be
limited except as otherwise set forth in the claims.
* * * * *
|
|
 |
|
 |
Description |
|
