Row Redundancy Block Architecture

1. Field of the Invention

The present invention generally relates to semiconductor memories and, more particularly, to row redundancy match detection with interchanged self-timing generator, a sample wordline enable generator with row redundancy match detection, and row redundancy control circuits for Dynamic Random Access Memory (DRAM) redundancy block architecture.

2. Background Description

Redundancy block architecture for DRAMs, proposed by H. L. Kalter et al. in "A 50-ns 16-Mb DRAM with 10-ns Data Rate and On-chip ECC", IEEE J. Solid-State Circuits, vol. 25, no. 5, October 1990, features flexible row redundancy replacement. This flexibility is advantageous because it allows redundancy elements and redundancy match-detection-decoders to be effectively used without increasing their number. This reduces the redundancy design space overhead significantly, especially for high density DRAMs such as 256-Mb DRAMs and beyond, while allowing good fault repairability.

In contrast to conventional intra-block redundancy replacement, redundancy block architecture is required to handle more redundancy match-detection-decoders "simultaneously", although the total number of redundancy requirements is significantly less. This is because of the flexibility of this architecture. It is very important to arrange the decoders effectively and to enable a fast and reliable redundancy match detection to avoid a potential speed and area penalty for the actual implementation.

It is therefore an object of the present invention to provide a redundancy block architecture which uses a row redundancy control circuit arrangement that effectively reduces design space.

It is a further object of the invention to provide a fast and reliable redundancy match detection by means of a NOR type redundancy match detection with an interchanged self-timing generator.

It is another object of the invention to provide a sample wordline enable (SWLE) generator which allows SWLE to set the sample wordline (SWL) at a time when wordline enable (WLE) sets the wordline (WL) regardless of normal or redundancy mode, while tracking the delay for the redundancy match detection.

The present invention is related to the invention disclosed in U.S. Pat. No. 5,517,442 to Kirihata et al., the subject matter of which is incorporated herein by reference. According to the present invention, row redundancy control circuits are arranged parallel to word direction and are arranged at the bottom of the redundancy block. This architecture change makes it possible to effectively lay out the redundancy control block by introducing (1) split-global-bus shared with local row redundancy wires, (2) half-length-one-way row redundancy-wordline-enable-signal wires which allows space saving, and (3) distributed wordline enable decoders designed to take advantage of the saved space. According to another aspect of the invention, an illegal normal/redundancy access problem caused by the address versus timing skew has been solved. The timing necessary for this detection is given locally by using its adjacent redundancy match detection. This allows the circuit to operate completely as an address driven circuit, resulting in fast and reliable redundancy match detection. In addition, a sample wordline enable signal (SWLE) is generated by using row redundancy match detection. One two-input OR gate allows the time at which SWLE sets sample wordline (SWL) to be the same as the time at which wordline enable (WLE) signal sets wordline (WL). The time at which SWLE sets SWL remains consistent regardless of mode, eliminating the existing reliability concern. This two-input OR gate combined with row redundancy match detection works as an ideal sample wordline enable generator.

In general, the purposes of the present invention are the following:

Provide an ideal arrangement for redundancy-match-detection-decoders to reduce the design space.

Allow fast and reliable redundancy-match-detection.

Realize a fast and reliable sample wordline enable generator using redundancy-match-detection post decoders.

The foregoing and other objects, aspects and advantages will be better understood from the following detailed description of a preferred embodiment of the invention with reference to the drawings, in which:

FIG. 1 is a block diagram of a 256-Mb DRAM;

FIG. 2 is a block diagram of a single 16-Mb unit of the 256-Mb DRAM shown in FIG. 1;

FIG. 3 is a block and logic diagram of the sixteen row redundancy (RRDN) control circuits and a 16-input NOR gate used in the 16-Mb unit shown in FIG. 2;

FIG. 4 is a block diagram of the split global bus architecture used in the preferred embodiment of the invention;

FIG. 5 is a block diagram of a 16-Mb unit used in the architecture shown in FIG. 4 showing the locations of the wordlines (WLs) and redundant wordlines (RWLs);

FIG. 6 is a block and logic diagram showing the RRDN and wiring arrangement of the 16-Mb unit of FIG. 5;

FIG. 7 is wiring diagram showing a RWL arrangement for multi-WL replacement according to the invention;

FIG. 8 is a block and logic diagram of a row redundancy (RRDN) control circuit;

FIG. 9 is a timing diagram showing the operation of the RRDN of FIG. 8 in normal and redundancy modes;

FIG. 10 is a circuit diagram of the fuse latch (FLAT) circuit;

FIG. 11 is a timing diagram showing the operation of the FLAT on power-up;

FIG. 12 is a table showing the operation of the FLAT for the conditions of existing and blown fuses;

FIG. 13 is a logic diagram of an AND type redundancy match detection circuit;

FIG. 14 is a timing diagram showing the operation of the circuit of FIG. 13;

FIG. 15 is a logic diagram, similar to FIG. 8, of a NOR type redundancy match detection circuit;

FIG. 16 is a timing diagram showing the operation of the circuit of FIG. 15;

FIG. 17 is a logic diagram of a NOR type redundancy match detection circuit with interchanged self-timing generator;

FIG. 18 is a timing diagram showing the operation of the circuit of FIG. 17;

FIG. 19 is a logic diagram of an existing sample wordline enable (SWLE) generator;

FIG. 20 is a timing diagram showing the operation of the SWLE generator of FIG. 19;

FIG. 21 is a logic diagram of the SWLE generator with row redundancy match detection;

FIG. 22 is a timing diagram showing the operation of the SWLE generator of FIG. 21;

FIG. 23 is a block diagram of the split global bus architecture of FIG. 4 modified with one sample wordline (SWL) in the center of the chip and shared with other 16-Mb units; and

FIG. 24 is a block diagram of the split global bus architecture of FIG. 4 modified with one SWL in each 16-Mb unit.

Referring now to the drawings, and more particularly to FIG. 1, there is shown a 256-Mb DRAM 10. This architecture is comprised of sixteen 16-Mb units 11, one of which is shown in more detail in FIG. 2. Each 16-Mb unit is divided into two arrays, the primary array 21 containing 256 wordlines (WLs) and a redundant array 22 containing 16 redundant wordlines (RWLs). The redundant array 22 allows the repair of up to sixteen faults, using sixteen row redundancy (RRDN) control circuits as shown in more detail in FIG. 3, where a wordline (WL) and redundancy wordlines (RWLi) are set by a wordline enable (WLE) signal and the corresponding redundancy wordline enable signal (RWLEi), respectively. In a standby state, WLE and RWLE are both low, and the N nodes are all high. In a normal mode, all the N nodes go low, enabling WLE (i.e., WLE goes high). In a redundancy mode, one RWLE goes high, and the corresponding node N stays at a high level, not enabling WLE (i.e., WLE stays at a low level). The more detailed operations to control WLE and RWLEs are described below.

Each row redundancy (RRDN) control circuit 31 compares the row addresses (ADDRs) with its programmed fuse states and activates either corresponding node Ni or RWLEi signal. A sixteen input NOR gate decoder 32 receives the outputs from the nodes Ni. When at least one address in an RRDNi control circuit 31 does not match its programmed fuse state, the signal at the corresponding node Ni falls, while keeping the corresponding RWLEi at a low level. When all ADDRs match its RRDN programmed fuse states, the corresponding RWLEi signal rises, but a corresponding node Ni remains at high level. If all nodes Ni from RDDNs fall, WLE rises, which is a normal mode. If one node Ni remains high, WLE remains low, which is a redundancy mode. In this redundancy mode, the corresponding RWLEi rises, activating the corresponding redundancy wordline RWLi.

The architecture requires one WLE signal, sixteen RWLE<0:15> signals, and sixteen N<0:15> signals for the outputs from RRDNs 31. Twenty address wires for possible 1K replacement are also required. In general, RRDN circuit design space is determined by the number of wires, because devices can be designed under the wires and the wiring space is more than sufficient. The architecture requires at least fifty-two wires (in addition to the necessary control signal and power lines), resulting in a minimum 156 .mu.m (52 .mu.m.times.3) design space, assuming a wiring pitch of 3 .mu.m. According to a first aspect of the invention, a solution is provided which effectively reduces additional design space for the redundancy control circuits to only 27 .mu.m.

FIG. 4 shows a split global bus architecture as disclosed in U.S. Pat. No. 5,517,442 to Kirihata et al. which is assumed for the present invention. In this architecture, each top/bottom 128 -Mb half of the chip has an independent global address bus 41 and 42, respectively. The 256-Mb DRAM consists of sixteen 16-Mb units 43, one of which is shown in more detail in FIG. 5. The 16-Mb unit consists of 8K wordlines (Wls). Any defective element in the unit is replaceable by any of sixteen redundancy wordlines (RWL<0:15>) in the redundancy block 52. The WLs and RWLs are alternatively driven from the left and right wordline drivers. Sixteen row redundancy (RRDN<0:15>) control circuits 53 are arranged at the bottom of each 16-Mb unit.

FIG. 6 shows the detailed RRDN circuit 53 and wiring arrangement for the invention. More particularly, the invention provides a row redundancy control circuit arrangement which effectively reduces design space. This reduction is achieved by (1) split global-address-bus shared with local row-redundancy wires, (2) half-length-one-way redundancy-enable wires for double-sided row decoder, and (3) distributed wordline enable decodes.

(1) Split global address bus shared with row redundancy local wires: The local address wires are shared with level two metal (M2) split global bus lines as described in U.S. Pat. No. 5,517,442 to Kirihata et al. RRDN circuits are laid out under the wires. This eliminates the necessity for the twenty address wires and devices for RRDN circuits, while shortening the address input wires for the RRDN circuits from the global bus lines.

(2) Half-length-one-way redundancy-enable wires for double-sided row decoder: The eight row redundancy (RRDN<0:7>) circuits designed in the left half unit controls the corresponding eight redundancy wordlines (RWL<0:7>) which are driven from the left drivers. The other eight row redundancy (RRDN<8:15>) circuits in the right half unit control the corresponding eight redundancy wordlines (RWL<8:15>) which are driven from the right drivers. This allows the maximum number of parallel redundancy wordline enable (RWLE<0:15>) wires to be halved.

(3) Distributed wordline enable decoders: The 16-input NOR decoders (32 in FIG. 3) which generate the signal WLE are distributed. The wires (N<0:15>) for the inputs to the distributed NOR gates are laid out in the unused space for the redundant wordline enable (RWLE<0:15>) lines. This results in the design space reduction for sixteen (N<0:15>) wires. The WLE signal is connected to both side WL drivers.

This description assumes a single WL replacement; however, the invention could be applied for a multi-wordline replacement. FIG. 7 shows a RWL arrangement for the multi-wordline replacement. If one of four WLs are defective, they are replaced with four corresponding RWLs. In order to allow up to a possible sixteen replacements, sixty-four RWLs are laid out in the redundancy block. Four consecutive RWLs act as a group and are used to replace four consecutive WLs when one of them is defective. Unlike the single RWL replacement, four consecutive redundant wordlines (RWL<4i:4i+3>) are controlled by one corresponding redundant word line enable (RWLE<i>) signal and are driven from the same side of the unit. Each adjacent four consecutive RWL group are driven from the other side of the unit.

FIG. 8 shows a detailed circuit of the redundancy match detection decoder (RRDN) 31 in FIG. 3. It consists of fuse latches (FLATs) 81 and a NOR gate decoder 82. As shown in FIG. 9, signal FADDR rises if the address ADDR does not match the programmed fuse state. As described above, the node N falls in normal mode, because at least one FADDR goes high. This disables RWLE activation even when signal SET rises. In a redundancy mode, the node N remains high, activating RWLE output and AND gate 83 when SET rises.

FIG. 10 shows a detail of the fuse latch (FLAT) in FIG. 9, which acts as an address/fuse comparator. The complementary metal oxide semiconductor (CMOS) latch is comprised of a pair of cross-coupled inverters 101 and 102 providing complementary outputs (via inverter 103) to transfer gates 104 and 105. The latch is set in chip power-up phase by signals FPUP and FPUN at the gates of PFET 106 and NFET 107, respectively, as shown in the timing diagram of FIG. 11. ADDR and ADDR are true and complement row addresses, respectively. Both ADDR and ADDR are low in a standby state. Either ADDR or ADDR rises when RAS falls (i.e., ADDR rises when the address is "0" and ADDR rises when the address is "1"). If a fuse 108 exists ("0" State), FADDR follows ADDR. FADDR, on the other hand, follows ADDR if the fuse is blown ("1" state).

The following discussion assumes that a programmed address is a "0" if the fuse is not blown and a "1" if the fuse is blown. FADDR accordingly rises when the address does not match the programmed fuse state (address/fuse unmatch detection). If ADDR and ADDR are swapped, FADDR rises when the address matches the programmed fuse state (address/fuse match detection). One FLAT is required for each address. In order to allow 1K possible replacement per array, ten FLATs are required.

According to another aspect of the invention, there is provided a row redundancy match detection circuit with interchanged self-timing generator that eliminates an illegal normal/redundancy access problem caused by the address versus time skew. In general, there are two approaches for the redundancy match detection; AND and NOR type. The match detection decoder illustrated in FIG. 8 is the NOR type. Conventional AND type and NOR type decoder operations are discussed next. FIG. 13 shows a row redundancy (RRDN) match detection circuit using AND gates 132, where a signal FDDR is an output from a fuse latch circuit (FLAT) 131. As shown in FIG. 14, signal FADDR rises if the corresponding address ADDR matches the programmed fuse state. A redundancy match detection RDMD<i> decodes all its FADDRs and controls its corresponding redundancy-wordline-enable signal RWLE<i>. Signal RWLE<i> will rise only when all its FADDRs are high (all input ADDRs match the programmed fuse states for its corresponding FLATs). A wordline-enable signal (WLE) from NOR gate 133, which activates a normal wordline (WL), will be enabled only if all RWLEs remain low. If one RWLE<i> rises, the corresponding redundancy-wordline will be enabled. In this case, the normal will not be enabled, because one RWLE<i> is high when the signal SET falls.

In this AND redundancy match detection, the timing for RWLE<i> is automatically generated by the ADDR transition (FADDRs are generated by ADDRs). The signal SET, however, must set the signal WLE. In order to avoid an illegal WLE activation, the time that SET falls must be later than the time that RWLE<i> rises. Although the time that SET falls could be determined according to the simulation results, it must be sufficiently delayed to avoid illegal WLE activation, resulting in a slow operation.

FIG. 15 shows a redundancy match detection using NOR gates 152. As shown in FIG. 16, signal FADDR rises if its corresponding ADDR does not match the programmed fuse state. RDMD<i> decodes all its FADDRs and controls its corresponding RWLE<i>. In a normal mode, all nodes N<i> fall when signals ADDR arrive, activating WLE. A node N<i> remains high only when all its FADDRs remain low (i.e., all input ADDRs match the programmed fuse states). In this case, the corresponding RWLE<i> AND gate 154 will be enabled when the signal SET rises. WLE output by NOR gate 153 remains low in the redundancy mode, because the N<i> remains high, disabling the normal mode.

In this NOR type redundance match detection, the timing for WLE is automatically generated by FADDRs (note that FADDRs are generated by ADDRs). The signal SET, however, must set the signal RWLE<i>. In order to avoid an illegal RWLE<i> activation, the time that SET rises must be later than the time that a node N<i> falls. Although the time that SET rises would be determined according to the simulation results, it must be sufficiently delayed in order to avoid illegal RWLE<i> activation, resulting in slow operation.

There is a requirement to design a fast and reliable row redundancy match detection. The existing detections described above require timing for either WLE or RWLE, resulting in an access speed penalty. The access speed penalty would be increased for the high density DRAMs (e.g., 256-Mb DRAMs), because the DRAMs must handle many row redundancy (RRDN) contact circuits and the SET and SET timing must be sufficiently delayed for all RRDN circuits. The invention allows a fast and reliable redundancy match detection by means of a NOR type redundancy match detection with an interchanged self-timing generator.

FIG. 17 shows a redundancy match detection using NOR gates 172, where N<i> generated by RDMD<i> is interchanged with the N<i+l> (which is generated by its adjacent RDMD<i+i>. N<i> is a timing signal for setting RWLE<i+I>. N<i+l> is a timing signal for setting RWLE<i>. Note that RDMD<i> and RWLE<i+i> cannot detect the redundancy mode simultaneously and cannot activate two RWLE<i> and RWLE<i+i>, because only one redundancy wordline at a time can be accessed in one array. As shown in FIG. 18, FADDR rises if the corresponding ADDR does not match the programmed fuse state. A NOR RDMD<i> decodes all its FADDRs and controls its corresponding RWLE<i>. In a normal mode, all nodes N<i> fall when FADDRs arrive, activating WLE. A node N<i> remains high only when all its FADDRs remain low (i.e., all input ADDRs for the corresponding FLATs match the programmed fuse states). In this case, N<i+l> generated by the adjacent RDMD<i+1> falls, activating the corresponding RWLE<i>. WLE remains low in the redundancy mode, because the N<i> remains high, disabling the normal mode.

The redundancy match detection is accomplished completely as an address driven circuit. No special timing generator is necessary. The adjacent RDMD<i+i> acts as a timing generator. The timing skew between N<i> and N<i+1> would be very small, because RDMD<i+l> uses the exact same layout of the RDMD<i> and is implemented near RDMD<i+1>. Almost no additional margin is required, resulting in a very fast and reliable redundancy match detection. According to another aspect of the invention, a sample wordline enable (SWLE) generator is provided which allows SWLE to set sample wordline (SWL) at a time when wordline enable (WLE) signal sets the wordline (WL) regardless of normal or redundancy mode, while tracking the delay for the redundancy match detection. As before, the following discussion assumes a redundancy array architecture which allows the repair of up to sixteen faults in a 16-Mb DRAM, using sixteen row redundancy (RDDN<0:15>) control circuits as shown in FIG. 4. It is further assumed that wordline (WL), redundancy wordlines (RWL<0:15>), and sample wordline (SWL) are set by the wordline enable (WLE) signal, the redundancy wordline enable (RWLE<0:15>) signal, and the sample wordline enable (SWLE) signal, respectively.

Each RDDN compares the row addresses ADDR with its programmed fuse states, and activates either corresponding N or RWLE signals. When a RRDN detects a normal mode (at least one address does not match its programmed fuse state), N falls. When RRDN detects a redundancy mode (all ADDRs match their RRDN programmed fuse states), the corresponding RWLE rises. WLE rises only when all sixteen Ns fall, as shown in FIG. 3.

A sample wordline (SWL), which simulates WL rising/falling delays, gives an ideal timing for the bitline sense Operation. It is important to set SWL at a time that WL is set. FIG. 19 shows the existing SWLE generator to set SWL in the form of OR gate 191. With reference to FIG. 20, when either WLE or RWLE rises, SWLE rises to set SWL. There is, however, a concern. The time at which SWLE sets SWL depends on whether the chip is in normal mode or redundancy mode. This SWLE timing change may cause a potential timing problem depending on the access mode, resulting in reliability problems. The concern becomes significant if one SWLE is shared with other units to save power dissipation, because the time (a) or (a') at which SWLE sets SWL is completely independent of the time (b) at which WLE sets WL (or the time (b') at which RWLE sets RWL) in other units, resulting in insufficient signal development time for some cells.

FIG. 21 shows the SWLE generator according to the invention, where two 8-input NOR gates 211 and 212 and one AND gate 213 compose the 16-input NOR WLE generator. A SWLE signal is given by using row redundancy match detection. One two-input OR gate 214 allows the time at which SWLE sets SWL to be the same as the time at which the WLE signal sets the wordline (WL). The time at which SWLE sets SWL remains consistent regardless of mode (normal or redundancy), eliminating the existing reliability concern. This two-input OR combined with row redundancy match detection works as an ideal sample wordline enable generator. As shown in FIG. 22, the time at which SWLE sets SWL is the same as the time at which WLE sets WL. The signal SWLE can accordingly set SWL without using RWLEs even in a redundancy mode, because either node 0 or 1 rises. Note that only one RRDN can detect a redundancy mode in each RAS cycle. Consequently, the time at which SWLE sets SWL is consistent regardless of the mode (normal or redundancy), while tracking the delay for the redundancy detection. The time at which RWLE set the corresponding RWL in the redundancy mode is earlier than the time at which WLE sets WL in the normal mode because RWLE is directly activated by the corresponding RDDN circuit. SWLE sets SWL later than RWLE (the set time of SWLE is the same as the set time of WLE in the normal mode). This should not be a problem (or may be an advantage), because this increases the signal development time for the row redundancy cells.

One SWL 231 is designed in the center of the chip and shared with other 16-Mb units as shown in FIG. 23. Sixteen RRDNs 232 in the unit are used for the SWLEt (top) and SWLEb (bottom) generators using the circuit described above. Note that these units are the slowest redundancy match detection points in the chip. SWL is set when either SLWEt or SWLEb is activated.

One SWL 241 is designed in each 16-Mb unit as shown in FIG. 24. SWLE is generated by using sixteen RDDNs 242 in each unit. This architecture has an additional advantage that SWLE sets SWL, while tracking the address wiring delay.

While the invention has been described in terms of preferred embodiments, those skilled in the art will recognize that the invention can be practiced with modification within the spirit and scope of the appended claims. It will be appreciated that the purposes of the present invention have been met; specifically to provide an ideal arrangement for redundancy-match-detection-decoders to reduce the design space, to allow fast and reliable redundancy-match-detection, and to realize a fast and reliable sample wordline enable generator using redundancy-match-detection post decoders.