This application claims foreign priority benefits under 35 U.S.C. §119 to co-pending German patent application number DE 10 2006 029 169.7-55, filed 24 Jun. 2006. This related patent application is herein incorporated by reference in its entirety.
1. Field of the Invention
The present invention relates to memory products and the operation of the same.
2. Description of the Related Art
In customary memories such as DRAM chips, the large number of binary memory cells is combined into a plurality of separate areas. Normally, the cells in each memory area form a matrix of rows and columns, and each column has an associated local read/write amplifier, subsequently referred to as “local amplifier” for short. Each of these amplifiers is connected to all the memory cells in the relevant column by means of an associated bit line. Each row can be selectively addressed by activating an associated row selection line (“word line”). The relevant activation signal is derived from a row address in a word line decoder (row decoder). The effect of activation is that each cell on the relevant line communicates its memory content to the local amplifier associated with the relevant column, which amplifier is then locked in a state in which it produces an amplified signal which represents the binary value of the stored data item.
Along each matrix row there extend one or more local data lines, and each data line is jointly associated with a plurality of local amplifiers. Each local amplifier has an individually associated column selection switch of its own which can be closed via a column selection signal in order to connect the amplifier to the associated local data line. The local data line can be connected via a line switch to an associated master data line which is common to a plurality of memory areas. Each master data line is routed to an associated “master amplifier”, which for its part is connected to an associated data connection for the purpose of outputting data which have been read and for the purpose of inputting data which is to be written.
By closing a column selection switch, the respective associated local data line and the further data path are used to set up the connection to one of the aforementioned data connections. In read mode, each closed column selection switch is used to transfer the latched data item from the addressed memory cell, amplified in the associated local amplifier, to the master amplifier, and in write mode, the data item which is input on the master amplifier is transferred via the closed column selection switch to the associated local amplifier, which then either maintains its previous latched state (when the write data item corresponds to the previous data item) or is changed over to the complementary latched state (when the write data item is different than the previous data item).
To transfer a read data item effectively from the local amplifier to the master amplifier in read mode and to latch a write data item effectively from the associated data line into the local amplifier in write mode, the associated column selection switch must respectively remain closed for a certain minimum time. When there is a given potential difference between the two defined logic levels which are intended to explicitly represent the binary data “0” and “1” on the data lines, this minimum time for the read mode is shorter the more powerful the local amplifier. However, it is not advisable to design the local amplifiers with a high gain factor in order to make said minimum time as short as possible. Powerful local amplifiers would each require a large amount of integration surface area, so that it would be a problem to accommodate them at such narrow intervals as are required by the column spacing of the memory cell matrix.
To keep down the circuit complexity, it is advantageous and also customary for the pulse duration and hence the closed time of column selection switches, subsequently referred to as “column selection time”, to be proportioned equally for read mode and write mode. The minimum duration is prescribed by design features of the memory chip, such as the gain of the local amplifiers and the charging time constants of the respective data line paths used. Each selected column selection switch is closed by a column selection pulse which, in response to an internal read or write command, is applied to a control line routed to the relevant switch and keeps this switch closed until the end of the pulse.
Like any digital circuit, memories such as DRAM chips are also subject to a clock control, where a superordinate clock signal CLK, comprising rising and falling pulse edges in regular succession, forms the time normal for the sequence of a multiplicity of individual operations and control processes. Thus, the timing of the succession of a series of read or write access operations to memory cells in the same matrix row, that is to say the “access clock”, is also controlled using this time normal. In this context, the period of the access clock, that is to say the time from one access operation to the next, is defined by a particular number of (whole or half) periods TC of the superordinate clock signal CLK. That is to say that this time and hence also the maximum available total time for the duration TS of a column selection pulse is proportional to the period duration TC of the clock signal CLK or inversely proportional to the clock frequency Fc=1/Tc.
A memory should be able to operate correctly over a defined range of clock frequencies fc. Therefore, there is a need to design an arrangement for activating the column selection in a memory chip such that it operates satisfactorily within a wide range of clock frequencies.
According to an embodiment is accordingly a memory which contains at least one matrix of memory cells which is organized into rows and columns, and whose operation is clock-controlled on the basis of a basic clock signal at the frequency fc and in which a chosen memory cell within an addressed matrix row is accessed by closing an addressed column selection switch which is associated with the matrix column containing the chosen memory cell in order to set up a connection for transferring a data bit between this cell and a data path. The embodiment provides a pulse generator which is started by a column selection command in the read mode and in the write mode of the memory chip in order to produce a column selection pulse which closes the addressed column selection switch and keeps it closed for the duration of this pulse. The pulse generator contains a first pulse timer for prescribing a fixed time Tf for the length Ts of the column selection pulse and a second pulse timer for prescribing a frequency-dependent time Tv, which is proportional to the clock signal period Tc=1/fc, for the length of the column selection pulse.
According to an embodiment, the pulse generator is in a form such that it has the first pulse timer take effect when the clock frequency fc is lower than a chosen threshold value fTH, and otherwise has the second pulse timer take effect.
Memories are known in which the length of the control pulse which closes the column selection switches addressed, or in other words the column selection time, is of fixed proportions. On the other hand, memory chips are also known in which the column selection time is variable and is proportioned so as to be inversely proportional to the respective clock frequency. To date, however, it has not been proposed or even merely suggested that a way be provided of changing between the two types of proportioning.
Embodiments of the invention is based on the insight that fixed proportioning of the column selection time may give rise to particular problems when the clock frequency fc exceeds a certain limit and hence the duration from the start to the end of a read or write cycle is below a certain length. In this context, within each cycle, the pause after each control pulse may end up so short that there is no longer sufficient time available to produce certain charge states for terminating the cycle. A further insight is that clock-dependent proportioning of the column selection time may give rise to other problems if the clock frequency fc is so low that the column selection pulse lasts much longer than is necessary for successfully reading and writing. This is undesirable for reasons of power consumption and susceptibility to interference. The reason for this is that the longer the column selection time the higher the power consumption and the greater the probability of interference signals being injected into the data transfer path too.
The controllability of the column selection time allows a memory chip to be operated over a wide range of clock frequencies to reduce the aforementioned problems.
The features of embodiments will become clear from the following description, taking in conjunction with the accompanying drawings. It is to be noted, however, that the accompanying drawings illustrate only typical embodiments and are, therefore, not to be considered limiting of the scope of the invention. It may admit other equally effective embodiments.
In the description below, the two logic potentials are denoted by H and L, where H is the “high” potential, which also represents the binary or logic value “1”, and L is the “low” potential, which is negative in comparison with H in the circuit example shown and represents the binary or logic value “0”.
Each memory cell in the matrix 10 is designed in the same manner as the memory cell 20 shown. It contains a capacitance 21 which forms the actual memory element and whose charge state represents the binary or data value “1” or “0” (uncharged). One side of the storage capacitor 21 is at fixed potential, at L- potential in the case shown, and the other side is connected via the channel of a selection transistor 22 designed as an N-FET to one of the two wires in an associated two-wire bit line BL. The gate of the selection transistor 22 is connected to a word line WL which is associated with all the cells in the same matrix row. In the usual vernacular, one of the wires is referred to as the “true” bit line wire BLt and the other is referred to as the “complement” bit line wire BLc. In the case illustrated, the selection transistor 22 is connected to the True-bit line wire BLt, just like the selection transistors on all the other cells on the same word line WL (the manner of connection usually changes from word line to word line).
The bit line wires BLt and BLc have a local amplifier 32 with a symmetrical input and a symmetrical output connected to them. The local amplifier 32 contains a first transistor pair, comprising two p-channel field effect transistors P1 and P2, and a second transistor pair, comprising two n-channel field effect transistors N1 and N2. The source electrodes of the P-FETs P1 and P2 are coupled together at a circuit point to which a first bias voltage potential VSP can be supplied. The source electrodes of the N-FETs N1 and N2 are coupled together at a circuit point to which a second bias voltage potential VSN can be supplied. The drain electrodes of the transistors P1 and N1 and the gate electrodes of the transistors P2 and N2 are connected to the bit line wire BLt and can also be connected via a first path of a two-pole column selection switch 33 to a first wire LDt in a local data line LD. Similarly, drain electrodes of the transistors P2 and N2 and the gate electrodes of the transistors P1 and N1 are connected to the bit line wire BLc and can also be connected via the second path of the column selection switch 33 to a second wire LDc in the local data line LD.
Further details from the circuit arrangement shown in
First,
During the quiescent state, the signals VSN and VSP are also kept at M-level, so that the local amplifier 32 is in the floating state. In the quiescent state, the column selection switch 33 is also open (i.e. off or nonconductive), so that the local amplifier 32 is isolated from the local data line LD. The open state of the line switch 43 means that the local data line LD is for its part isolated from the master data line MD. While the column selection switch 33 is open, a further precharge switch 51 on the master data line remains closed in order to keep the two wires MDt and MDc at a precharge potential equal to the H-level.
To prepare cell access, first of all, at time t1, an activation command ACT is produced, which renders the signal PRE inactive in order to open the precharge switches 31 and 41 and hence to isolate all the bit lines BL and all the local data lines LD from the precharge potential M and to close the line switch 43, so that both wires LDt and LDc in the local data line LD change to the H precharge potential of the master data lines MD. Roughly at the same time, the word line WLin a row x selected by a row address is activated, i.e. is lifted to H level, which it reaches after a certain charging time at time t2, so that the selection transistors 22 in all the memory cells 20 in the relevant row are turned on. The word line activation is effected by means of an associated word line driver 23 which is selected by a row decoder on the basis of a decoded row address x and is activated by the activation command ACT.
When the selection transistor 22 has been turned on, the bit line wire BLt receives the potential which represents the previous charge state of the cell 21 and hence the binary value of the stored data item. As an example,
A short time later, after a safety time has elapsed at time t3, if the change in potential on the bit line wire BLt has reached a degree which can be clearly discriminated from noise phenomena, the signal VSN is switched to L level and the signal VSP is switched to H level. This turns on the local amplifier 32 by virtue of it now receiving its full supply voltage and its operating current from the sources of the two signals. If a reduction in potential has taken place on BLt, as shown in
The process described above is executed simultaneously on all the columns in the addressed row, so that at time t4 the local amplifiers associated with the other columns (not shown in
To this end, an internal read command RD is given at a time t5 and at the same time the column selection switch 33 associated with the addressed column y is closed, specifically by activating a control line CSLassociated with this switch for the duration of a column selection pulse CS. The column selection pulse CS is produced in a pulse generator 60, where it is triggered by the read command RD, and is routed via a column decoder 13 to the control line CSLselected on the basis of the column address y. The precharge switch 51 on the master data line MD is opened for the duration of the column selection pulse CS, so that the H precharge potential is decoupled from the master data line wires MDt and MDc for this duration. This is symbolized in
When the column selection switch 33 is closed at time t5, the potential difference H-L of the bit line wires BLt, BLc which has been “separated” in the manner described is coupled to the local data line wires LDt, LDc which are connected to the master data line wires MDt, MDc via the closed line switch 43.
To transfer the data item latched on the local amplifier 32 to the master amplifier 52, a line switch 54 which is situated between the master data line MD and the amplifier 52 is also closed for the duration of the column selection pulse CS. For the data transfer from the local amplifier 32 to the master amplifier 52, it is necessary to overcome the previous charge states of the local data line LD and of the master data line MD until the resultant potential difference on the input of the master amplifier 52 is sufficient to toggle this amplifier to a state which corresponds to the data item (indicated in
The duration TS of the column selection pulse CS for the read mode therefore needs to be at least as long as TR, i.e. the column selection switch 33 cannot be opened again before time t6. In the case shown, the column selection pulse CS is terminated at a somewhat later time t7.
At the end of the column selection pulse CS, that is to say as soon as the local amplifier 32 has been isolated from the data transfer path LD, MD again, the precharge switch 51 is closed again and the line switch 54 is opened again, so that both wires MDt, MDc of the master data line MD and both wires MDt, LDc of the local data line LD attempt to attain the H precharge potential again. When this potential state has been reached, a further memory cell in the same row can be read while the word line WLcontinues to remain activated and all the local amplifiers remain on. To this end, a further read command RD can be given at a time t8, which follows t7 at a sufficient interval, and this in turn produces a column selection pulse CS, but this is put onto the control line of another column selection switch in response to a new column address. The cycle described for the time period t5-t8 can be repeated a plurality of times, each time using a read command RD at time t8, with a different column address being applied each time in order to apply the column selection pulse CS for closing a respective different column selection switch (“fast-page” mode).
After one or more read cycles on the same matrix row, the entire read operation can be terminated and the quiescent state which prevailed before time t1 restored. To this end, a precharge command PR is given at time t8 at the end of the last read cycle, which activates the precharge signal PRE to H again, deactivates the word line WL to L, puts the supply potentials VSN and VSP for the local amplifier 32 (and all other local amplifiers in the same matrix area) back to M potential, opens the line switch 43 again and puts the bit line wires BLt, BLc and the local data line wires LDt, LDc back to their precharge potential M.
The write cycle is started by an internal write command WR at a time t5′. Before time t5′, that connection of the master amplifier 52 which is associated with the master data line wire MDt was set to H potential and the other connection of the master amplifier 52, which is associated with the master data line wire MDc, was set to L potential, in line with the “1” to be written. The write command WR at time t5′ starts the column selection pulse CS, so that the line switch 54 on the master amplifier 52 is closed. As a result, MDc and LDc strive for L potential, while MDt and LDt remain at H potential. Since the appearance of the column selection pulse CS also causes the column selection switch 33 to close, the previous state of the local amplifier 32 is changed over as soon as the potential difference between the wires LDt and LDc reaches the switching threshold which is required for this (indicated in
The duration TS of the column selection pulse CS for the write mode therefore needs to be at least as long as TW, i.e. the column selection switch 33 cannot be opened again before time t6′. In the case shown, the column selection pulse CS is terminated at a somewhat later time T7′.
From the very start of the column selection pulse at time t5′, the closed switch 33 is used to pull the bit line wire BLt and hence also the potential V21 on the storage capacitor 21 in the cell 20 from L to H, while the bit line wire BLc is pulled from H to L. The charging current required for this is introduced, when the local amplifier 32 has toggled, both by this amplifier and by the master amplifier 52, so that the charge reversal on the bit line wires BLt and BLc occurs relatively quickly so long as the column selection pulse CS lasts.
As soon as the column selection pulse CS ends, that is to say from time t7′, the further charge reversal on the bit line BL and on the cell capacitor 21 occurs at a slower speed. That is to say that the cell charge reversal time TU from the start of when the column selection switch 33 closes to time t9′, at which the charge on the cell capacitor 21 is completely reversed, is longer the shorter the time for which the column selection switch is closed (that is to say the “column selection time” TS). It is equally true that the cell charge reversal time TU is shorter the longer TS is.
A further write cycle on another cell in the addressed matrix row can be started before the actual end t9′ of the actual cell charge reversal time TU, however, e.g. at a time t8′. In this context, the word line WLcontinues to remain activated, and all the local amplifiers remain on. It is thus possible to write to further memory cells in the same row in relatively brief succession by repeating the cycle running over the time period t5′-t8′ a plurality of times, in each case starting with an internal read command WR, the column address y being changed from cycle to cycle in order to apply the column selection pulse CS to a different column selection switch each time.
If the write operation is intended to be terminated by a precharge command PR, in order to set the quiescent state, which prevailed before activation at time t1′ again, this precharge command cannot be given until after the cell charge reversal time TU after the start of the last column selection pulse CS has elapsed. For this reason, the specifications for a memory chip also contain a fixed absolute time preset TWR for the time period from the start of the last column selection pulse for a write operation to the output of the precharge command (Write Recovery Time). The actual cell charge reversal time TU must never be longer than TWR.
The precharge command PR activates the precharge signal PRE to H again, the word line WL is deactivated to L, the supply potentials VSN and VSP for the local amplifier 32 (and for all other local amplifiers in the same matrix area) are set to M potential again, the line switch 43 is opened again and the bit line wires BLt, BLc and also the local data line wires LDt, LDc are brought to their precharge potential M again.
For the time control of the read and write cycles described above, the memory chip contains a control device which produces the necessary signals in the desired chronology under the influence of a clock signal CLK. Thus, the time period t5′-t8′ and also the equally long time period t5-t8 from the start to the end of a read cycle (
TZ=n*TC/2=n/2fc.
As already mentioned further above, the pause period TZ-TS from the end of the column selection pulse CS (time t7 or t7′) to the end of the access cycle (time t8 or t8′) should not fall below a certain minimum measure TPmin, because otherwise the control line CSLand particularly also the data lines LD and MD would not regenerate sufficiently by the end of the cycle. When the pulse duration TS is given fixed proportions, this pause becomes shorter the higher the clock frequency fc. At high clock frequencies, the pause period can therefore drop below the desired minimum measure.
This danger can be reduced to some degree if it is ensured that the pulse duration TS is inversely proportional to the clock frequency, that is to say changes to the same degree and in the same direction as the cycle time TZ. Although the aforementioned pause period TZ-TS then likewise changes to the same degree, it remains at a finite value. On the other hand, however, it has been found that clock-dependent proportioning of the column selection time TS is disadvantageous for the read mode if the clock frequency is relatively low. The relatively long column selection time in this case can result in leakage phenomena and the injection of interference becoming noticeable, which corrupt the data item read. There is therefore not just a minimum time TSmin to be observed for the column selection pulse CS and a minimum time to be observed for the subsequent pulse pause but also a maximum time TSmax for the pulse CS, which should not be exceeded if at all possible.
The general design of the pulse generator 60 which is shown in
The first pulse timer 70 is designed such that its output reverts to L level again as soon as a fixed time period TF has elapsed after the time of triggering. This fixed time period TF is proportioned such that it is not shorter than TSmin. The minimum time TSmin is a variable which is dependent on the layout and on the operating voltage values and also on other operating conditions of the memory chip and can be ascertained for each chip type empirically or through simulation. In this context, it would be necessary to take account of the worst case which is to be expected, i.e. the one in which the time period for changing over the master amplifier in read mode and the time period for changing over the local amplifier in write mode are longest. In one advantageous embodiment, TF is equal to TSmin (as shown as an example in
The second pulse timer 80 is designed such that its output reverts back to L level when a time period TV since the time of triggering has elapsed which is variable and which is dependent on the frequency fC of the clock signal CLK which is supplied to this pulse timer for the purpose of control. This control is effected such that the variable time TV is proportional to the period duration TC of the clock signal CLK, to be more precise equal to a number k>n of half clock signal periods:
Tv=k*TC/2.
In this case, the proportionality factor k is chosen such that TV becomes equal to TF when the clock frequency fC decreases to such an extent that the difference TZ-TPmin becomes equal to TF.
As soon as the output of one of the two pulse timers reverts to L level, the output of the AND gate 62 also changes back to L level, which terminates the column selection pulse CS. The duration TS of the pulse CS is thus equal to the respective shorter of the two time periods TF and TV. That is to say that when the clock frequency fC is higher than a threshold value fTH at which TV=TF, the duration TS of the column selection pulse CS is equal to the frequency-dependent value TV. Otherwise, that is to say at lower clock frequencies fC≦fTH, the duration TS of the column selection pulse has the fixed value TF.
As soon as the active edge of a read command RD or of a write command WR appears and is routed to the pulse timer 70 by the OR gate 61, the flip-flop 72 is set, so that its Q output changes from L level to H level. When the delay time TF has elapsed, the flip-flop 72 is reset again by the output edge of the delay device 71. A pulse of length TF thus appears at the Q output of the flip-flop.
The embodiment of the second pulse timer 80 which is shown in
As soon as the active edge of a read command RD or of a write command WR appears and is routed to the pulse timer 80 by the OR gate 61, the flip-flop 83 is set, so that its Q output changes from L level to H level. At the same time, the flip-flop 81 is also set in order to use its Q output to activate the counter 82 so that this counter counts the clock pulse edges which appear at time intervals of TC/2. As soon as k edges have been counted, that is to say after a time period Tv=k*TC/2, a decoded count output of the counter produces an edge which is supplied to the reset input R of the flip-flop 83 in order to reset this flip-flop. Hence, a pulse of length Tv=k*TC/2 appears at the Q output of the flip-flop 83. At the same time as the flip-flop 83, the flip-flop 81 and the counter 82 are also reset, so that the pulse generator 60 is ready for fresh triggering by a subsequent read or write command.
The circuits shown in
For the exemplary embodiment described, it has been assumed that both the rising and the falling edges of the clock signal can be used as “active” edges. For this reason, the numbers n and k have been defined as a number of half clock signal periods. If clock edges with just one particular polarity (rising or falling) may be active, however, the numbers n and k need to be defined as a number of whole clock signal periods.
The minimum value for the factor k is 1, and the minimum value for the factor n is accordingly equal to 2. A cycle duration TZ equal to one clock signal period (that is to say equal to two half clock signal periods) is real for ordinary fast memory chips. In this case, the pulse timer for the variable pulse duration may be a simple RS flip-flop whose set input S receives the active edge of a write or read command which coincides with an active clock edge and whose reset input receives the subsequent active clock edge.
As already mentioned further above, an inventive pulse generator allowing the column selection time TS to be changed between a value which is independent of the clock frequency and a value which is dependent on the clock frequency can advantageously be designed such that the memory chip can be tested inexpensively.
The usual course of a memory test is to operate the memory chip under conditions which are as similar as possible to the conditions of the subsequent use mode. In this context, the operating response is tested under the extreme conditions of the specification. One of these extreme conditions is the maximum value fCmax of the specified range of clock frequencies. The inventive pulse generator for the column selection pulses will, when operated at the maximum frequency fCmax, supply the clock-dependent pulse duration, specifically the smallest value of its envisaged value range.
At maximum clock frequency fCmax and hence with the shortest column selection time TSmin, the cell charge reversal time TU has the maximum value TUmax. This operating condition is therefore the most critical for the TWR specification. A memory test could therefore involve performing a write cycle for each memory cell, which reverses the charge on the relevant cell, at the maximum clock frequency fCmax and setting the time of the precharge command PR after the end of the cycle in each case so that the time period from the start of the column selection pulse to this time is shorter than the specified value TWR by a small measure ε. In a subsequent read mode, it can then be verified whether writing has taken place correctly on all memory cells. If so, the memory chip operates satisfactorily with the TWR specification, i.e. the cell charge reversal time TU is shorter than TWR for each memory cell.
A test of this kind is very cost-intensive because it is necessary to use a rapidly operating test appliance, which is very expensive. The pulse generator is provided with a switching device which allows the memory test to be performed at a much lower clock frequency than fCmax and still to disclose errors which would occur at maximum clock frequency. This switching device can be activated by a special test mode command in order to define the duration of the column selection pulse independently of the actual clock frequency fC as being a time value which is equal to the value TSmin which would be produced in use mode on the basis of frequency at maximum clock frequency fCmax.
This can be checked by using a precharge command PRE precisely at the end of the time period TUmax to deactivate the addressed word line, that is to say to stop any further charging possibility for the addressed memory cell, then reading the relevant memory cell again and comparing the data item which has been read with the data item written. If there is a match, it can be assumed that the write path to the relevant memory cell is in order, so that the write cycle ought to work satisfactorily even in use mode at maximum clock frequency fCmax within the specified range and with the TWR specification. If there is no match, the charge reversal on the memory cell has not been able to be extensive enough, for example because the associated local amplifier is too weak.
The time period TUmax which is needed when the memory chip is operating properly and for a given duration TSmin of the column selection pulse CS before charge reversal on the cell capacitor 21 is complete can be discovered experimentally or through simulation.
The precharge command at the end of TUmax is also used to put the supply potentials VSN and VSP for the local amplifier 32 (and for all other local amplifiers in the same matrix area) back to M potential, and likewise the bit line wires BLt, BLc and the local data line wires LDt, LDc. The line switch 43 is opened again, so that the wires MDt and MDc in the master data line also temporarily change to M potential if the column selection pulse lasts beyond the time t9″ and hence the precharge switch 51 on the master data line MD is still open. The master data line wires MDt and MDc are not brought to bear H precharge potential until the column selection pulse CS ends at a later time.
A pulse generator as shown in
The pulse generators 60a and 60b shown in
In the case of the pulse generator 60b shown in
It should also be mentioned that the switches shown in the figures, which are shown symbolically as mechanical switches, are in reality naturally electronic switches, advantageously formed by field effect transistors.
The preceding description describes advantageous exemplary embodiments. The features disclosed therein and the claims and the drawings can, therefore, be useful for realizing various embodiments, both individually and in any combination. While the foregoing is directed to specific embodiments, other and further embodiments may be devised without departing from the basic scope, the scope being determined by the claims that follow.
Number | Date | Country | Kind |
---|---|---|---|
102006029169.7-55 | Jun 2006 | DE | national |