Integrated memory core and memory interface circuit

Abstract
A memory device comprises a first and second integrated circuit dies. The first integrated circuit die comprises a memory core as well as a first interface circuit. The first interface circuit permits full access to the memory cells (e.g., reading, writing, activating, pre-charging and refreshing operations to the memory cells). The second integrated circuit die comprises a second interface that interfaces the memory core, via the first interface circuit, an external bus, such as a synchronous interface to an external bus. A technique combines memory core integrated circuit dies with interface integrated circuit dies to configure a memory device. A speed test on the memory core integrated circuit dies is conducted, and the interface integrated circuit die is electrically coupled to the memory core integrated circuit die based on the speed of the memory core integrated circuit die.
Description
BACKGROUND OF THE INVENTION

1. Field of the Invention


The present invention is directed toward the field of building custom memory systems cost-effectively for a wide range of markets.


2. Art Background


Dynamic Random Access Memory (DRAM) is the most popular type of volatile memory and is widely used in a number of different markets. The popularity of DRAMs is mostly due to their cost-effectiveness (Mb/$). The PC main memory market has traditionally been the largest consumer of DRAMs. However, in recent times, other important markets have adopted DRAMs. A report published by De Dios and Associates showed that in 2004, the PC main memory market consumed only 50% of the total DRAM bits.


Several of the non-PC markets use specialty or legacy memories. Specialty memory is typically memory that is not used by the PC main memory but is memory that is designed for one or more niche markets. For example, the PC graphics market uses GDDR (Graphics Dual Data Rate) DRAM. Similarly, some segments of the network infrastructure market use FCRAM (Fast Cycle RAM) or RLDRAM (Reduced Latency DRAM). Legacy memory is typically memory that was used in the past but is not used in that particular market segment now. For example, SDRAM (Synchronous DRAM) was used for PC main memory from ˜1997 to ˜2001 but is no longer used today for PC main memory. Instead, most cellular phones and handheld (or mobile) devices use SDRAM today.


Bringing a new DRAM architecture into the market requires significant investment of time and money. For example, it typically takes 4 years for JEDEC to approve a new DRAM architecture. DRAM makers must then spend hundreds of millions of dollars to productize the new architecture. Unless the investment is amortized over an extremely large number of devices, the cost of the new devices will be high. In addition, the DRAM makers have optimized their manufacturing flow for high volumes. Any deviation from the norm disrupts the flow. This is the reason why specialty and legacy memory typically carry a price premium over memory used by the PC main memory market (which is usually referred to as commodity memory).


Given the time and money required to bring a new DRAM architecture into the market, it is obvious that the industry does not have the luxury of being able to define a DRAM architecture that exclusively meets the needs of the smaller markets for DRAMs. For example, it is difficult for the DRAM makers to cost-effectively produce a DRAM that perfectly meets the needs of the cell phone market. Therefore, it is even more difficult for each cell phone maker (e.g. Nokia or Motorola) to design DRAMs tailor made for its phones. So, cell phone designers are forced to choose the DRAM architecture that is least objectionable from their perspective. Hence the selection of SDRAM for cell phones.


This situation will become even worse in the future. Most analyst projections show that not only will DRAM usage expand rapidly into newer markets but also that the DRAM bit consumption growth rate will be higher for non-PC markets. The needs of these markets are quite different from the needs of the PC main memory market. Clearly, there is a need in the market for a way to quickly and cost-effectively build custom memory that is tailor made for a customer's exact needs.


SUMMARY OF THE INVENTION

A memory device comprises a first integrated circuit die. The first integrated circuit die comprises a memory core, with a plurality of memory cells, and a first interface circuit for accessing the memory cells of the memory core. For example, the first interface circuit provides reading, writing, activating, pre-charging and refreshing operations to the memory cells. A second integrated circuit die, electrically coupled to the first integrated circuit die, comprises a second interface circuit for accessing the memory core via the first interface circuit and for interfacing the memory core to an external circuit. For example, the second interface circuit may comprise a synchronous interface to an external bus. As such, the memory device has two separate die: one for the memory core and a second as an external interface.


In one embodiment, the memory core includes a plurality of memory banks for partitioning the memory cells. A multiplexer, coupled to the memory banks, selects data from one or more of the memory banks. The multiplexer is located generally near an edge of the first integrated circuit die. The first integrated circuit die further comprises data input/output (“I/O”) pads and a plurality of bond wires that couple the multiplexer to the I/O pads. The I/O pads are located essentially adjacent to the multiplexer near an edge of the first integrated circuit die so as to minimize distance of the bond wires.


In another embodiment, a distributed-bank architecture is used to configure the memory device. For this embodiment, the memory cells are partitioned into memory banks. The memory banks comprise a plurality of sub-arrays across the physical sections of the memory core such that a physical section of the memory cells comprises a plurality of sub-arrays associated with different memory banks. A multiplexer selects a memory bank from a physical section.


In some embodiment, the first and second integrated circuit dies are housed in separate packages. In other embodiments, the first and second integrated circuit dies are housed in the same package.





BRIEF DESCRIPTION OF THE DRAWINGS


FIG. 1 is a block diagram illustrating a conventional DRAM chip.



FIG. 2 illustrates a typical organization of a 4-bank modern SDRAM.



FIG. 3 is a block diagram illustrating one embodiment of banks arranged in sub-arrays.



FIG. 4 illustrates a block diagram of an interface and multiple banks in a DRAM.



FIG. 5 is a block diagram illustrating a DRAM chip with an interface removed.



FIG. 6 illustrates one embodiment for a center bonded DRAM core chip.



FIG. 7 illustrates one embodiment for an edge bonded DRAM core chip.



FIG. 8 illustrates one embodiment for a concentrated-bank architecture.



FIG. 9 illustrates one embodiment for a distributed-bank architecture.



FIG. 10 illustrates one embodiment for a quadrant in a distributed-bank architecture DRAM core chip.



FIG. 11 is block diagram illustrating one embodiment of a distributed-bank architecture universal DRAM core chip.



FIG. 12 is a block diagram illustrating a distributed-bank architecture DRAM core chip configured to support DDR2 speeds.



FIG. 13 is a block diagram illustrating a distributed-bank architecture DRAM core chip configured to support DDR speeds and external bus widths from 17 to 32 bits.



FIG. 14 is a block diagram illustrating a distributed-bank architecture DRAM core chip configured to support DDR speeds and external bus widths from 9 to 16 bits.



FIG. 15 is a block diagram illustrating a distributed-bank architecture DRAM core chip configured to support DDR speeds and external bus widths from 1 to 8 bits.



FIG. 16 illustrates one embodiment for a portion of a DRAM core chip that includes a decoder for selecting a mode of operation.



FIG. 17
a is a block diagram illustrating the relationship between the internal data bus rate and the external data bus rate for 4n pre-fetching.



FIG. 17
b is a block diagram illustrating the relationship between the internal data bus rate and the external data bus rate for burst mode with a length of 4n.



FIG. 18 is a block diagram illustrating an example multi-chip memory implementation.



FIG. 19 illustrates techniques for stacking two DRAM core chips behind a single interface chip.





DETAILED DESCRIPTION

The disclosure of U.S. Provisional Patent Application Ser. No. 60/693,631, entitled “Methods and Apparatus for Integrating Multi-Chip Memory Devices”, filed on Jun. 24, 2005, is hereby expressly incorporated herein by reference.


By examining several different DRAM architectures, it is clear that the internal organizations of the DRAMs are quite similar. For example, a ×16 (×16 denotes the external data width) 256 Mb SDRAM, ×16 256 Mb DDR SDRAM, and ×16 256 Mb DDR2 SDRAM consist of:


Control logic block;


Address input register and decoder;


Memory arrays that store the data;


Data selection circuit (I/O gating);


Data read circuit; and


Data write circuit.


It is obvious that most of the blocks are common across all the three architectures. The main differences are in the control logic block (which implements the protocol, among other functions), in the width of the data that is accessed per column address, and in the data I/O section. These are usually considered part of the interface section of the DRAM while the rest of the circuits (address decoder, memory arrays, and data selection) are considered part of the memory core. The core timing parameters are typically specified in absolute units of time (seconds) rather than in terms of clock periods. For example, the Micron 256 Mb DDR2 SDRAM data sheet lists the following core timing parameters:














TABLE 1







Speed Grade
tRCD (ns)
tRP (ns)
tRC (ns)









−5E
15
15
55



−37E
15
15
55



−3
15
15
55



−3E
12
12
54










Similarly, the Micron 256 Mb DDR SDRAM data sheet identifies the following timing specifications:














TABLE 2







Speed Grade
tRCD (ns)
tRP (ns)
tRC (ns)









−75Z
20
20
65



−75
20
20
65










The Micron 256 Mb SDRAM data sheet discloses the following specifications:














TABLE 3







Speed Grade
tRCD (ns)
tRP (ns)
tRC (ns)









−7E
15
15
60



−75
20
20
66










So, even though the protocol and speed of SDRAM, DDR SDRAM, and DDR2 SDRAM are quite different, it is clear that the internal core or array of all these types of DRAMs has similar characteristics. In fact, we can go even further and observe that all synchronous DRAMs are composed of an asynchronous core and an interface that defines the protocol, synchronous operation, speed, and signaling. The memory core typically comprises ˜90%-95% of the total die area.


Current practice is to integrate the memory core and the interface onto a common die. The drawback with this approach is that a change in the protocol, speed, or signaling for example requires a re-design of the entire chip. This is usually very expensive and time consuming, and hence the inability to bring specialty or custom DRAMs to the market quickly and cost-effectively. One embodiment of the invention comprises a multi-chip implementation, wherein one or more DRAM core chips are attached to an interface chip. The interface chip sits between the host electronic system and the DRAM core chips. In other words, the interface chip can be thought of as a “wrapper” that surrounds the DRAM core chips. The partitioning of the conventional DRAM into DRAM core chip and interface chip should preferably be done is such a way that that the functions and circuits that are relatively constant across many different architectures are retained in the DRAM core chip while the functions and circuits that vary between the different architectures are moved to the interface chip.


The DRAM core chip can be designed to be suitable for a large number of markets (i.e. a “universal core”). The interface chip can now be designed to meet the exact needs of a market, and even the exact needs of individual customers in that market. To illustrate, the proposed solution enables the design of an interface chip to meet the exact needs of Nokia for the cell phone market and another interface chip to meet the exact needs of Motorola for the cell phone market.


In order to accommodate the needs of the different markets, the DRAM core chip must be capable of operating across a wide range of frequencies, be capable of supporting high data rates, and must be low cost. In one embodiment, the DRAM core chip is asynchronous, wide, and operates at its natural speed. For the case of modern DRAM cores, the natural speed is between 5 ns to 10 ns per column access, which is equivalent to 100 MHz to 200 MHz synchronous operation. That is, a modern DRAM core can keep up with an external memory bus or interface that runs at a speed from 100 MHz to 200 MHz. So, for the case of a synchronous DRAM that operates at 100 MHz to 200 MHz and is n-bits wide (1≦n≦32 typically), n bits can be fetched from the DRAM core once every clock cycle. In fact, this is how SDRAMs operate.


Newer synchronous DRAMs run at higher clock speeds. JEDEC defines the DDR SDRAM specification with external data rates of 200 MHz, 266 MHz, 333 MHz, and 400 MHz. An even newer specification called DDR2 SDRAM has been defined with external data rates of 400 MHz, 533 MHz, 667 MHz, and 800 MHz. Effort is currently underway in JEDEC to define a DDR3 SDRAM specification that spans data rates from 800 MHz to 1600 MHz. GDDR, GDDR2, and GDDR3 SDRAMs typically run faster than the DDR, DDR2, and DDR3 SDRAMs. However, even though the external data rate has been increasing quite rapidly, the speed of the DRAM core has not kept pace. In order to bridge the gap between the external data rate and the internal core speed, the DRAM industry has adopted a technique called “pre-fetching.”


Pre-fetching involves accessing more bits than the external data bus width on every column access. To illustrate, an n-bit wide DDR SDRAM accesses 2n bits every column access. This allows the external data bus to run at 200 MHz to 400 MHz while the internal memory core runs at 100 MHz to 200 MHz respectively. FIG. 1 is a block diagram illustrating a conventional DRAM chip. A DRAM chip 100 comprises a DRAM core 110, Internal Data Bus 120, DRAM interface 130 and External data bus 140. TABLE 4 shows the concept of pre-fetching for a DRAM chip.













TABLE 4






External
Internal
External
Internal



Data Bus
Data Bus
Data
Data


Protocol
140 Width
120 Width
Rate (MHz)
Rate (MHz)







SDRAM
N
n
 66-133
 66-133


DDR SDRAM
N
2n
200-400
100-200


DDR2 SDRAM
n
4n
400-800
100-200


DDR3 SDRAM
n
8n
 800-1600
100-200


(proposed)









This implies that the universal DRAM core chip must be sufficiently wide enough to support the data rates required by many different markets. Obviously there is a limit to how wide the universal DRAM core chip can be before it starts to negatively impact the cost of the chip. In general, if the width of the DRAM core chip is so large so as to make either the core chip or the interface chip pad limited (especially the core chip), the cost of this solution would be very high.


Modern DRAMs also feature multiple banks. A bank is a section of the DRAM core that can be accessed independently. The DRAM core is broken up into banks that can be active simultaneously. Within each bank, only one row can be open at any given time. Most DRAMs up to 512 Mb densities are organized into 4 banks. 1 Gb (and possibly, up to 4 Gb) DRAMs are organized into 8 banks but only 4 banks can be activated within a specific time window. This is dictated by power and thermal considerations. So, the universal DRAM core chip must be capable of supporting multiple banks.


Let us consider the internal organization of a ×16 256 Mb SDRAM. A ×16 256 Mb SDRAM may have 4 banks, each of which is 64 Mb. Each bank can be conceptualized as consisting of 16 sub-arrays, each sub-array being a 8192×512 matrix of memory cells. That is, each sub-array has 8192 or 8 k rows and 512 columns. So, when a bank is accessed, a particular row is accessed (activated) in each of the 16 sub-arrays in the bank. The row is determined by the row address. After the 16 rows are activated, a particular bit in each row is selected. The bit is specified by the column address. So, on each access to a bank, 16 bits are accessed.



FIG. 2 illustrates a typical organization of a 4-bank modern SDRAM. The memory cells are arranged into four banks: bank 0 (220), bank 1 (210), bank 2 (230) and bank 3 (240). Each bank contains P×Q×16 cells (e.g., P=8192 and Q=512 for a 256 Mb SDRAM). Each bank has associated word line drivers (275, 280, 285 and 290) and sense amplifiers (255, 260, 265 and 270). The banks are selected through use of MUX 250.


In one embodiment, the banks are organized in sub-arrays. FIG. 3 is a block diagram illustrating one embodiment of banks arranged in sub-arrays. For this embodiment, each bank has 16 sub-arrays (each sub-array being 8K×512) because the DRAM is organized as a ×16 memory.


Consider the internal organization of a ×16 256 Mb DDR SDRAM. The ×16 256 Mb DDR SDRAM is organized similar to the ×16 256 Mb SDRAM with some changes to the memory core. The more important changes to the core organization are:


Each bank has 32 sub-arrays; and


Each sub-array is now 8192×256 matrix (i.e. P=8192, Q=256). The reason for having 32 sub-arrays is that DDR SDRAM memory uses a pre-fetching of 2n. Since this is a ×16 DDR memory, 32 bits must be accessed from each bank for a read or write operation.


Note that pre-fetching can be done in a number of ways. Consider a memory array that is organized as a P×Q matrix that needs to support 2n pre-fetching. One approach is to divide the P×Q array into two arrays (i.e. two P×Q/2 arrays) and access both arrays in parallel, so that we get 2 bits per column address. Another approach is to not split the array but modify the column decoder so that 2 bits are selected for each column address (in other words, the least significant bit of the column address is not used). Some embodiments of the invention are described that use the first approach. However, the teachings of the present invention are applicable to different pre-fetching implementations.


Looking at the organization of a ×16 256 Mb DDR2 SDRAM, the ×16 256 Mb DDR2 SDRAM is organized similar to the ×16 256 Mb SDRAM (and the ×16 256 Mb DDR SDRAM). The following identify some of the changes to the memory core:


Each bank has 64 sub-arrays. Each sub-array is now a 8192×128 matrix (i.e. P=8192, Q=128). The reason for the 64 sub-arrays per bank is that DDR2 SDRAM uses a 4n pre-fetching. Since this is a ×16 DDR2 memory (n=16), 64 bits must be accessed from each bank for a read or write operation.


In all cases (SDRAM, DDR, DDR2), data bits to/from each bank are brought to a multiplexer/de-multiplexer (hereafter referred to as a MUX), which in turn is connected to the external DQ pins. This MUX is typically in the middle of the DRAM chip. FIG. 4 illustrates a block diagram of an interface and multiple banks in a DRAM. As shown in FIG. 4, bank 0 (410), bank 1 (420), bank 2 (430) and bank 3 (440) are accessed by interface 450. Note that for SDRAM, m=n, for DDR SDRAM, m=2n , DDR2 SDRAM, m=4n , and DDR3 SDRAM (proposed), m=8n , Also note that the data MUX is typically part of the interface.


This arrangement works well for a conventional DRAM since everything is on a single die. However, in one embodiment of the invention, the interface is on a separate die. If we were to just move the interface alone (to another die) without disturbing the memory core, then the number of I/O pads on both the memory core chip and the interface chip will become quite large, as shown in TABLE 5 below for a ×16 4-bank implementation. FIG. 5 is a block diagram illustrating a DRAM chip with an interface removed. For this embodiment, bank 0 (510), bank 1 (520), bank 2 (530) and bank 3 (540) are coupled to I/O pads 550, 560, 570 and 580, respectively. If we look at only the data pins and ignored the address, command, power and ground pins, we can see that the number of data signals that have to go off-chip between the DRAM core chip and the interface chip is 4m.














TABLE 5








External

Total Number of




Data Bus

Off-Chip Data



Protocol
Width (n)
m
Pins (4 m)





















SDRAM
16
16
64



DDR SDRAM
16
32
128



DDR2 SDRAM
16
64
256



DDR3 SDRAM
16
128
512










So, it is quite obvious that removing the interface without disturbing the rest of the memory core quickly leads to a very large number of off-chip connections, especially for wider external data bus widths and higher data speeds (because, the amount of pre-fetching will increase with higher data speeds). Under these conditions, either the DRAM core chip or the interface chip or both will become pad limited, which will increase the cost of the total solution.


In one embodiment, in order to reduce the number of off-chip connections between the DRAM core chip and the interface chip, some part or all of the multiplexing of the data I/O from the banks is done in the core chip itself.


One option is to route all the data bits from each bank to a central MUX, and then connect the other side of the MUX to off-chip drivers. This is quite similar to the current practice for center bonded DRAMs. FIG. 6 illustrates one embodiment for a center bonded DRAM core chip. For this example, integrated circuit 600 includes bank 0 (610), bank 1 (620), bank 2 (630) and bank 3 (640) coupled to MUX 650. MUX 650 is connected to substrate bonding pads 670 via bond wires 680 through I/O pads 660.


The drawback with this approach is that bond wires 680 that connect I/O pads 660 on the DRAM core die to the substrate bonding pads 670 become quite long. Long bond wires have significant inductance and limit the speed at which the memory chip can operate.


In another embodiment, edge bonding for the core chip is used. FIG. 7 illustrates one embodiment for an edge bonded DRAM core chip. For this embodiment, integrated circuit 700 includes bank 0 (710), bank 1 (720), bank 2 (730) and bank 3 (740). I/O pads 750, located on the silicon die 705, are connected to the substrate bonding pads 760 via bond wires 780.


If the DRAM core was organized with one bank per quadrant, then the number of data I/O pads will be equal to 4m as illustrated previously. The other option is to route the data bits from each bank to a centrally located MUX (as shown in FIG. 6) and then route the signals from the other side of the MUX to the periphery of the die. However, this means that the data signals will have to traverse the die twice—once from the bank to the central MUX and once from the central MUX to the periphery. This increases routing complexity, may possibly require an extra metal layer on the die (higher cost), and adds to the latency of the memory core.


In another embodiment, an inventive “Distributed-Bank” architecture is used. In this architecture, a bank is distributed (or spread) across all 4 quadrants instead of concentrating a bank in only one quadrant. Using this architecture, data MUXs, located in all 4 quadrants, select the appropriate bank, and the data signals corresponding to the selected bank can be easily routed to the periphery of the chip.



FIG. 8 illustrates one embodiment for a concentrated-bank architecture. For the purpose of this illustration, a ×16, 4-bank, 256 Mb DDR2 SDRAM core is used. However, any type of DRAM, with different external data widths, different number of banks, different density, and different amount of pre-fetching may be used without deviating from the spirit or scope of the invention. As previously shown, each bank (810, 820, 830 and 840) in a ×16, 4-bank, 256 Mb DDR2 SDRAM consists of 64 sub-arrays, with each sub-array organized as a 8192×128 array of memory cells.



FIG. 9 illustrates one embodiment for a distributed-bank architecture. For this embodiment, the DRAM core chip is divided into four quadrants (910, 920, 930 and 940). Each quadrant includes a portion of a bank.



FIG. 10 illustrates one embodiment for a quadrant in a distributed-bank architecture DRAM core chip. As discussed previously, there are 64 sub-arrays per quadrant with each sub-array being a 8192×128 matrix. Instead of assigning all 64 sub-arrays in a single quadrant to a single bank in the concentrated-bank architecture, there are 16 sub-arrays to each of the 4 banks within a single quadrant in the distributed-bank architecture. In the distributed-bank architecture embodiment, local data MUXs are located in each quadrant to select one of the four banks.



FIG. 11 is block diagram illustrating one embodiment of a distributed-bank architecture universal DRAM core chip. For this embodiment, the banks of DRAM cells are distributed among quadrant 1110, 1120, 1130 and 1140. An address decoder 1150, located in the center of the chip, controls word line drivers 1155, 1164, 1170 and 1176 in quadrants 1110, 1120, 1130 and 1140 respectively. The data from the bank sub-arrays are output to sense amplifiers (1157, 1166, 1172 and 1178) and input to the respective bank select MUXs (1160, 1168, 1174 and 1180). The data is then routed to data I/O pads 1162 located in the proximity for each of the quadrants.


Since 64 data bits are accessed from the core chip for every read or write operation, the interface chip may be designed or configured to act similar to a ×16 DDR2 SDRAM, ×8 DDR2 SDRAM, ×4 DDR2 SDRAM, ×2 DDR2 SDRAM, or ×1 DDR2 SDRAM. In fact, the interface chip may be designed to support any data width between ×1 and ×16 when operating in a 4n pre-fetch mode.


The distributed-bank architecture is flexible enough to support protocols like SDRAM, DDR SDRAM, DDR2 SDRAM, and DDR3 SDRAM. For example, the DRAM core chip shown in FIG. 11 may be configured or used as shown in FIG. 12 to support DDR2 speeds. FIG. 12 is a block diagram illustrating a distributed-bank architecture DRAM core chip configured to support DDR2 speeds. MUX 1210 selects 64 bits of data for one of the banks (1220, 1230, 1240 and 1250). MUX 1210 represents the data MUXs located in all four quadrants on the DRAM core chip.


The DRAM core chip shown in FIG. 11 may also be configured or used as shown in FIG. 13 to support DDR SDRAM speeds when it is operated in a 2n pre-fetch mode. FIG. 13 is a block diagram illustrating a distributed-bank architecture DRAM core chip configured to support DDR speeds and external bus widths from 17 to 32 bits. MUX 1310 selects 64 bits of data for one of the banks (1320, 1330, 1340 and 1350). The mode of operation shown in FIG. 13 may be used with the appropriate interface chip to support external data widths between (and inclusive) of 17 and 32 in a 2n pre-fetch mode. The same DRAM core chip may be used with the appropriate interface chip to support external data widths between (and inclusive) 9 and 16 in a 2n pre-fetch mode when operated as shown in FIG. 14. FIG. 14 is a block diagram illustrating a distributed-bank architecture DRAM core chip configured to support DDR speeds and external bus widths from 9 to 16 bits. For this embodiment, MUX 1410 selects 32 bits of data for one of the banks (1420, 1430, 1440 and 1450).


Also, the same DRAM core chip can be used with the appropriate interface chip to support external data widths between (and inclusive of) 1 and 8 in a 2n pre-fetch mode. FIG. 15 is a block diagram illustrating a distributed-bank architecture DRAM core chip configured to support DDR speeds and external bus widths from 1 to 8 bits. For this embodiment, MUX 1510 selects 16 bits of data for one of the banks (1520, 1530, 1540 and 1550).


From these architectures, the internal data bus width (the width of the bus between the DRAM core chip and the interface chip) may be configured to match the amount of pre-fetching required (which is determined by the external data rate) and the width of the external data bus. The external data bus is the bus from the interface chip to the ASIC or memory controller. The DRAM core chip as shown in FIG. 11 may be configured to support the following modes and requirements shown in TABLE 6.












TABLE 6





Internal Data
Pre-Fetching
Minimum External
Maximum External


Bus Width
Used
Data Bus Width
Data Bus Width


















64
8n
1
8



4n
1
16



2n
1
32



1n
1
64


32
8n
1
4



4n
1
8



2n
1
16



1n
1
32


16
8n
1
2



4n
1
4



2n
1
8



1n
1
16









Note that:


The proposed DDR3 SDRAM is an example of an 8n pre-fetch protocol;


DDR2 SDRAM is an example of a 4n pre-fetch protocol;


DDR SDRAM is an example of a 2n pre-fetch protocol; and


SDRAM is an example of a 1n pre-fetch protocol (i.e. no pre-fetching needed).


Again, for the DRAM core chip shown in FIG. 11, 3 modes of operation may be defined and a 2-bit binary code may be assigned to represent them as shown in TABLE 7.












TABLE 7







Mode of Operation (Mode[1:0])
Internal Data Bus Width









00
64



01
32



10
16



11
Undefined/Reserved










These two bits (Mode[1:0]) may be inputs to the DRAM core chip so that the internal data bus width is selected through external means. For example, the Mode[1:0] inputs to the core chip may be selected by means of fuses on the core chip or on the interface chip, by means of pull-up or pull-down resistors in the package of either chip (or in the common package) or on the printed circuit board, or may be driven by a register on the interface chip, or may be part of the address input to the core chip.


Let us assume that the Mode[1:0] inputs to the DRAM core chip are controlled by a register in the interface chip. FIG. 16 illustrates one embodiment for a portion of a DRAM core chip that includes a decoder for selecting a mode of operation. The decoder 1610 in the core chip is aware of the Mode[1:0] inputs as shown in FIG. 16.


Note that the embodiments disclosed below are based on the DRAM core chip shown in FIG. 11 only for the purpose of explaining the concept, and that the following embodiments are applicable to DRAM core chips of different densities, number of banks, internal organization, and number of sub-arrays. For the purpose of simplicity, only bank 0 is shown being accessed in the different modes of operation.


The mode decoder truth table is shown below in TABLE 8. In TABLE 8 below,


RA=Row Address


X=Don't Care


H=Asserted


L=Not Asserted















TABLE 8





Mode[1:0]
RA[14]
RA[13]
Bank0a_En
Bank0b_En
Bank0c_En
Bank0d_En







00
X
X
H
H
H
H


01
X
L
H
H
L
L



X
H
L
L
H
H


10
L
L
H
L
L
L



L
H
L
H
L
L



H
L
L
L
H
L



H
H
L
L
L
H









Based on the techniques of the present invention, a universal DRAM core chip, such as the embodiment shown in FIG. 11, may be configured to support a wide variety of data speeds and widths. For example, the core chip shown in FIG. 11 may support data rates up to 8*f MB/s, where f is the maximum clock rate (in MHz) at which the DRAM core can run in sync with the external data bus without using pre-fetching (i.e. using a pre-fetching of 1n ). For modern DRAM processes and designs, f is typically between 100 MHz and 200 MHz. So, the DRAM core chip shown in FIG. 11 supports maximum data rates between 800 MB/s and 1600 MB/s (1.6 GB/s).


In order to build custom memory cost-effectively, it is imperative that the DRAM core chip be used in a variety of markets. This will reduce the cost of the core chip because of economies of scale. Since the memory core is typically 90% to 95% of the silicon area, the overall cost may be lowered. Here, we can make two observations:


Some markets for the universal DRAM core chip value low power at the expense of performance (e.g. cell phones and other handheld devices) whereas other markets will sacrifice power to achieve higher speed (e.g. PC graphics and game console markets).


Semiconductor fabrication process is inherently statistical in nature. That is, if we fabricate a statistically significant number of identical chips, some of the chips will only be capable of operating below the target speed, some of the chip will be capable of operating at the target speed, and some of the chips will be capable of operating above the target speed. These are known in the industry as slow, typical, and fast parts respectively. Usually, the fast parts are sold at a price premium over the other parts while the slow parts are sold at lower prices compared to the typical parts.


However, slow parts typically consume less power than the typical parts, which in turn typically consume less power than the fast parts. So, if we can sort the DRAM core chips according to their maximum speeds of operation (usually called “speed binning”) before they are attached to the interface chips, we can:


Attach interface chips designed for low power markets (e.g. an SDRAM-like interface for the cell phone market) to the slow core parts;


Attach interface chips designed for the high performance/speed markets (e.g. a GDDR3-like interface for the game console market) to the fast core parts; and


Attach interface chips designed for markets sensitive to both power and performance (e.g. a DDR-like interface for the server market) to the typical core parts.


This allows us to maximize the ASP (average selling price or average sales price) of all the solutions since all the core chips have natural homes.


Speed binning of memory chips is typically done after it is packaged. Note that it is possible to do a simple speed sort of the memory chips at the wafer level itself. In order to do a speed sort or speed bin, we need to use ATE (automatic test equipment), also known as tester.


When we speed bin the DRAM core chips, we need to measure the time required for basic operations like Read, Write, Activate (open one or more pages), Precharge (close one or more pages), and Refresh. To satisfy this requirement, the DRAM core chips, defined by the present invention, are fully functional asynchronous DRAM chips capable of stand-alone operation. In other words, the DRAM core chips contain all the necessary circuits and capabilities needed to access the internal array used to store the data.


Memory makers, especially DRAM manufacturers, build redundancy into the memory core. For example, if the memory array is to be organized as P×Q (P rows and Q columns), the actual array is designed as (P+i)×(Q+j), where i and j are small compared to P and Q respectively. This allows the memory makers to replace up to i defective rows in the main array with the redundant rows, and up to j defective columns in the main array with the redundant columns. With the help of the redundant rows and columns, memory makers can increase the yield (i.e. the percentage of fully functional chips) to ≧90%. In a typical DRAM manufacturing flow, the individual dies on a wafer are tested at low speed and the partially functional dies (i.e. those with some defective rows and/or columns) are marked. The defective rows and/or columns on these marked dies are replaced with the redundant rows and/or columns respectively.


However, a die that uses the redundant rows and/or columns (because it had some defective rows and/or columns) will be slower than a die that does not use redundant rows and/or columns. This is due to the nature of how redundancy is built into the memory and how it is enabled. Therefore:


Interface chips that are designed for high performance markets may be attached to the DRAM core dies that do not have any defective rows and/or columns in the main array.


In another embodiment, the defective rows and/or columns of memory core dies are not replaced with the redundant rows and/or columns, but are configured to operate the memory core dies as (P/y)×(Q/z), where y and z are preferably powers of 2 (including 20=1). These DRAM core chips may then be attached to interface chips that are designed for high performance markets.


As we have seen previously, the DRAM makers use pre-fetching to support higher external data rates. For example, DDR2 SDRAM uses 4n pre-fetching. This means that for an n-bit wide external data bus, 4n data bits are accessed from the memory core for every read or write. In a conventional DRAM (where the memory core and the interface are on the same die), increasing the amount of pre-fetching increases the amount of metal interconnects on the die, which has a modest impact on the cost. In the invention described herein, increasing the amount of pre-fetching may make either the memory core chip or the interface chip or both pad limited. Being pad limited can increase the cost substantially.


Burst mode is another technique that can be used to increase the data rate of memory chips. In burst mode, the memory chip reads or writes multiple data bits per column address. For example, an n-bit wide (external data bus width) memory chip that is configured for a burst mode of 4n will access 4n bits from the memory core for a given column address. So this is quite similar to a 4n pre-fetch except that in burst mode, the same data wires are used. In other words, in a memory chip that supports 4n burst mode (but not 4n pre-fetching), the internal data bus between the memory core and the interface is only n-bits wide. Each line in the internal bus carries 4 data bits that are separated in time.


The difference between pre-fetching and burst mode is shown in FIGS. 17a and 17b. FIG. 17a is a block diagram illustrating the relationship between the internal data bus rate and the external data bus rate for 4n pre-fetching. Memory core 1710 is coupled to memory interface 1720 via internal data bus 1715 at 4n @ f1, Hz data rate. The memory interface 1720 is coupled to external data bus 1725, and under these conditions, the external data bus operates at a n @ 4*f2, Hz data rate. FIG. 17b is a block diagram illustrating the relationship between the internal data bus rate and the external data bus rate for burst mode with a length of 4n. Memory core 1730 is coupled to memory interface 1750 via internal data bus 1740 at n @ 4*f2, Hz data rate. The memory interface 1750 is coupled to external data bus 1760, and for the burst mode of operation, the external data bus operates at a n @ 4*f2, Hz data rate.


Typically, pre-fetching will provide higher external data rates than burst mode. However, burst mode does not increase the amount of off-chip connections between the core chip and the interface chip. So, in some embodiments, it is preferable to design the DRAM core chip of this invention with burst mode capability.


As mentioned previously, one of the aspects of this invention is the ability to test and speed bin the memory core chips and then attach the appropriate interface chips. Testing and speed binning of the DRAM core chip is usually done on a tester. This requires the core chip to have sufficiently strong output drivers to drive the inputs of the tester, which are usually some distance (several inches) from the outputs of the core chip. However, in the normal mode of operation, the inputs of the interface chip will be much closer (<1″) to the outputs of the core chip. So, it is not necessary to have strong output drivers in the core chip in the normal mode of operation. In order to satisfy both requirements, in some embodiments the DRAM core chip preferably has output drivers whose strength or drive capability is adjustable. For example, the core chip may have, by default, normal strength output drivers that are capable of driving signals across some distance to the inputs of the tester. However, when an interface chip is attached to the core chip, a signal from the interface chip decreases the drive strength of the core chip's output drivers.


In some embodiments, the output drivers of interface chip that interface with the DRAM core chip have similar adjustable drive strength capability. This allows testing of the interface chips separately prior to attaching them to the core chips. Note that the adjustable drive strength drivers are not necessarily required on the interface chip on the pins that interface to the electronic host system. However, it is preferable to have the adjustable strength capability on these drivers as well so that the drive strength can be tailored to the requirements of the system or external world. As with the DRAM core chips, the strength of the output drivers on the interface chip that communicate with the core chip are preferably controlled by a signal from the core chip.



FIG. 18 is a block diagram illustrating an example multi-chip memory implementation. The solution includes a DRAM core chip 1810 and an interface chip 1820. In some embodiments, the main characteristics of the DRAM core chip of this invention are:


Asynchronous or synchronous DRAM that is capable of stand-alone operation;


A plurality of DRAM core chips may be coupled together with one or more interface chips;


A plurality of interface chips may be coupled together with one or more DRAM core chips;


The interface on the DRAM core chip may include a custom and/or industry standard interface;


Has address inputs (bank address, row address, column address—row and column address can be on separate inputs or multiplexed on same pins);


Has command inputs like address strobes, read/write, output enable, and data masks);


Has control inputs that determine mode of operation—examples are inputs that determine the width of the internal data bus (bus between the memory core chip and interface chip) and inputs that determine the strength of the output drivers;


Has control outputs that determine some aspect of the functions performed by the interface chip;


Internal data bus width ≧external data bus width (bus from interface chip to memory controller or ASIC);


Optional burst mode capability;


Adjustable drive strength on output drivers;


Capable of well-defined standard operations like Read, Write, Activate, Precharge, and Refresh that can be clearly characterized in terms of speed; and


May be tested, burnt in, and speed binned independently (i.e. in stand-alone mode).


In some embodiments, the main characteristics of the interface chip of this invention are:


Implements the protocol used by the memory controller;


Interface chip implements an industry standard protocol like DDR SDRAM, DDR2 SDRAM, GDDR2 SDRAM, etc.;


Interface chip implements an industry standard protocol with custom extensions (e.g. GDDR2 SDRAM with extensions as specified by mutual agreement with one or more customers);


Interface chip implements a fully custom protocol as specified by one or more customers or a fully custom protocol developed in-house;


Interface chip operates as a transformer to convert protocols from the external interface to the interface of the DRAM core chip (e.g., synchronous to asynchronous and asynchronous to synchronous);


Interface chip determines the signaling used by the external interface;


For example, single ended, pseudo-differential, fully differential;


For example, push-pull outputs, open drain/collector outputs;


For example, asynchronous, synchronous, source synchronous, SerDes-like where clock is encoded/embedded in the data stream;


Determines the width of the external data bus;


Determines the speed of operation of the memory chip (by memory chip, we mean the combination of the DRAM core chip and the interface chip);


Determines the pin out of the memory chip as seen by the external world;


Allows the pin out of the memory chip to better match the pin out of the ASIC/memory controller to reduce the board routing complexity;


Implements special or custom functions and modes of operation;


Example, special power management functions and operating modes; and


Example, special error detection and error correction capabilities, as well as other types of redundancy capabilities and functions.


The DRAM core chip and the interface chip of this invention may be attached together in a number of different ways:


One or more DRAM core chip dies and one or more interface chip dies may be electrically connected to each other and the whole combination be put into a single package (e.g., a single DRAM core chip die with a single interface chip die, multiple DRAM core chip dies with a single interface chip die, or a single DRAM core chip die with multiple interface chip dies).


The core chip die may be put in a separate package, and then the interface chip die may then be electrically attached to the package containing the core chip die;


The interface chip die may be put in a separate package, and then the core chip die may then be electrically attached to the package containing the interface chip die;


The core chip die can be put in a separate package; the interface chip die may be put in a separate package; and the two packages can be electrically attached to each other;


Multiple DRAM core chip dies may be put in a separate package; the interface chip die may be put in a separate package; and the two packages can be electrically attached to each other;


A DRAM core chip die may be put in a separate package; multiple interface chip dies may be put in a separate package; and the two packages can be electrically attached to each other;


The DRAM core chip die and the interface chip die may be electrically attached in any way without deviating from the spirit or scope of the invention.


One aspect of this invention is that a multi-chip DRAM that is built according to this invention, might have higher cost than a traditional DRAM, especially if the DRAM core chip die and the interface chip die were packaged separately and then attached to each other. This is due to the cost associated with the extra package. One way to ameliorate this is to put multiple DRAM core chip dies in a single package. For the purpose of this discussion, we shall consider putting two DRAM core chip dies in a single package, each die being a 256 Mb density device. A typical DRAM manufacturing process might have the following sequences after the wafer has been fully processed:


DRAM dies on a wafer are tested at low speed and dies with defective rows and/or columns are marked;


The defective rows and/or columns are replaced with redundant rows and/or columns;


Wafer is diced into individual dies, which are then packaged;


Packaged parts are tested for functionality—parts damaged by the packaging process are eliminated;


Tested packaged parts undergo long term burn in to eliminate infant mortality parts; and


Burnt in parts are optionally tested again for functionality and shipped.


So, if we place two 256 Mb DRAM core dies in a single package, the following 3 bins may be generated after the parts have been packaged and burnt-in:


Bin A—both the DRAM core chip dies are functional, so the total capacity is 512 Mb


Bin B—only one of the DRAM core chip dies is functional, so the total capacity is 256 Mb


Bin C—neither of the DRAM core chip dies is functional, so the total capacity is 0 Mb


The bin C parts should be discarded. We can now use the bin B parts for those markets and/or customers who require only 256 Mb devices. For example, a handheld device manufacturer might require only a 256 Mb DRAM. So, the bin B parts can be attached to the interface chips designed for this manufacturer. Other markets and/or manufacturers might require 512 Mb devices. For example, a network router manufacturer might need 512 Mb DRAMs. So, we can use bin A parts for this manufacturer by attaching the appropriate interface chips to the bin A parts. This concept can be extended to cover more than two DRAM core chip dies in a single package as well as DRAM core chip dies of all densities.


As disclosed above, some embodiments cover the idea of attaching multiple DRAM core chips to a single interface chip. The core chips may be attached to the interface chip in a number of different ways. Some of these ways are described below and in FIG. 19. Again, for the purpose of illustration, we shall assume that two 4-bank 256 Mb DRAM core chips (FIG. 11) are attached to the interface chip. Each of the core chips has a 64-bit wide data bus designed to connect to the interface chip. Note that the idea explained below can be applied to DRAM core chips with different number of banks, density, data bus width, etc.


In one method (1910), the row addresses are used to select the DRAM core chip. In a second method (1920), bank addresses are used to select DRAM core chips. In a third method (1930), the column addresses are used to select the DRAM core chip.


The two DRAM core chips are attached to the interface chip such that only one of the core chips is accessible at any given time. That is, the two core chips look to the electronic host system as a single 512 Mb DRAM with 4 banks. This implies that the interface chip will use the row address to select one or the other core chip.


The two DRAM core chips are attached to the interface such that both of the core chips are accessible at any given time, and that the two chips look to the external world electronic host system as a single 512 Mb DRAM with 8 banks. This implies that the interface chip will use the bank address to select one or the other core chip.


The two DRAM core chips are attached to the interface chip such that both of the core chips are accessible at any given time, and that the two chips look to the electronic host system as a single 512 Mb DRAM with 4 banks. The interface chip uses the column address to select one or the other core chip. Note that in other embodiments an interface chip always accesses both the core chips in parallel, so that the data bus between the core chips and the interface chip becomes 128-bits wide. For this embodiment, the interface chip doubles the external data rate. In other words, the amount of pre-fetching has been doubled.


In other embodiments the multi-chip solution is configured such that the attachment of the DRAM core chips to the interface chip is programmable. For example, the customer may choose between one of the three methods listed above to attach two 4-bank, 64-bit wide, 256 Mb DRAM core chips to an interface chip by programmable means. These means include using fuses on the interface chip or the core chips, pull-up or pull-down resistors on the package substrates or printed circuit board, or by means of a register on the interface chip or core chips. However, any way of attaching the DRAM core chips to the interface chip may be accomplished without deviating from the spirit or scope of the invention.


Other embodiments of the invention include building redundant memory systems by attaching multiple DRAM core chips to an interface chip. For example, when more than one core chip is attached to an interface chip, redundancy is added by several means including:


Storing identical copies of the data in corresponding locations of each core chip when data is written to the memory by the memory controller; when the data is read back by the memory controller, the interface chip can read the multiple copies of the data from the different core chips and select the correct copy and transmit it to the memory controller. The correct copy can be determined by means like majority voting, and/or by the use of parity or ECC bits;


Using (n+m) bits to store n data bits.


Another aspect of this invention is placing memory core chips of different types behind a common interface chip. For example, we can place any combination of DRAM core, SRAM (Static Random Access Memory), and Flash chips behind a common interface chip. Multi-Chip Package (MCP) memory solutions are fairly common in the cell phone and handheld markets today. The issue with current MCP solutions is that each of these memories (DRAM, SRAM, Flash) has different interfaces, which complicates the design of the memory controller, the packaging, and the board routing. Placing any possible combination of DRAM core chip, SRAM, and Flash behind a common interface chip simplifies the memory controller design since the idiosyncrasies of each of these memory types is hidden from the memory controller. In addition, the board routing is simplified.


Another aspect of this invention is placing a large and slow memory as well as a smaller and faster memory behind a common interface chip and using the faster memory as a cache for the slower memory. For example, an SRAM chip might be used as the cache for a DRAM core chip or a DRAM core chip can be used as the cache for a Flash chip. The cache management logic may be built into the interface chip so that the cache is transparent to the memory controller. Alternately, the cache may be made visible to the memory controller and managed by the memory controller. Let us consider the case of an interface chip that has been designed to interface to one or more DRAM core chips and an SRAM chip. The SRAM chip can be used to cache the rows in the DRAM core chips that were recently opened. The SRAM may be used to cache the entire contents of the recently opened rows in the DRAM or cache part of the contents of the recently opened rows in the DRAM. The properties of the cache (associatively of the cache lines, mapping between DRAM rows and SRAM cache lines, etc.) may be determined by programming certain registers in the interface chip. By storing data that has a high likelihood of being accessed in the near future, system performance is improved.


Using a cache also allows the interface chip to do speculative pre-fetching of data from the DRAM core chip (and storing it in the SRAM cache chip), which again improves system performance. In addition, the interface chip may operate with SRAM chips (that are used as caches) of different densities. This allows the same interface chip to be used across several different segments within a market. For example, a DRAM/SRAM combination memory device may includes a common interface where the DRAM capacity is 512 Mb and the SRAM capacity ranges from 0 to 32 Mb. This allows a DRAM supplier to ship the 512 Mb DRAM+32 Mb SRAM combination in the high performance segment of the market, ship a 512 Mb DRAM+8 Mb SRAM combination in the mainstream segment of the market, and ship a 512 Mb DRAM (no SRAM cache) device in the value segment of the market.


The techniques of the present invention are not just applicable to DRAM. As such, the DRAM core chip need not necessarily be a trench-capacitor or stacked-capacitor device. The present invention is applicable to a variety of memory technologies like MRAM (Magnetic RAM), FRAM (Ferro-electric RAM), Ovonics memory, molecular memory (e.g. memory technology developed by ZettaCore), carbon nanotube memory (e.g. memory technology developed by Nantero Inc.), etc.


Another aspect of this invention is that it can be used with DRAM core chips that have different architectures like FCRAM (Fast Cycle RAM), RLDRAM (Reduced Latency DRAM), ESDRAM (Enhanced SDRAM).


Some embodiments of invention allow the use of a common memory core across a wide range of markets while varying the interface of the memory core according to the market and customer needs. It also allows the interface to be kept constant while changing the memory core behind the interface to address the needs of the different segments within a market.


Although the present invention has been described in terms of specific exemplary embodiments, it will be appreciated that various modifications and alterations might be made by those skilled in the art without departing from the spirit and scope of the invention.

Claims
  • 1. A memory device for use with an external circuit, the memory device comprising: at least one first integrated circuit die including,a memory core comprising a plurality of memory cells, anda first interface circuit including a first interface for accessing the memory cells of the memory core and configuring a data rate for transferring data between the memory cells and the first interface circuit; andat least one second integrated circuit die, electrically coupled to the first integrated circuit die, comprising a second interface for accessing the memory core via the first interface circuit and for interfacing the memory core to the external circuit.
  • 2. The memory device as set forth in claim 1, further comprising: a plurality of the first integrated circuit dies.
  • 3. The memory device as set forth in claim 1, further comprising: a plurality of the second integrated circuit dies.
  • 4. A memory device for use with an external circuit. the memory device comprising: at least one first integrated circuit die including,a memory core comprising a plurality of memory cells, anda first interface circuit including a first interface for accessing the memory cells of the memory core;at least one second integrated circuit die, electrically coupled to the first integrated circuit die, comprising a second interface for accessing the memory core via the first interface circuit and for interfacing the memory core to the external circuit;a first package housing the at least one first integrated circuit die; anda second package housing the at least one second integrated circuit die.
  • 5. The memory device as set forth in claim 1, further comprising: a single package housing the first and second integrated circuit dies.
  • 6. A memory device for use with an external circuit, the memory device comprising: at least one first integrated circuit die including,a memory core comprising a plurality of memory cells, anda first interface circuit including a first interface for accessing the memory cells of the memory core;a plurality of second integrated circuit dies, electrically coupled to the first integrated circuit die, comprising a second interface for accessing the memory core via the first interface circuit and for interfacing the memory core to the external circuit; anda single package housing the at least one first integrated circuit die and housing the plurality of second integrated circuit dies.
  • 7. A memory device for use with an external circuit, the memory device comprising: a plurality of first integrated circuit dies including,a memory core comprising a plurality of memory cells, anda first interface circuit including a first interface for accessing the memory cells of the memory core;at least one second integrated circuit die, electrically coupled to the plurality of first integrated circuit dies, comprising a second interface for accessing the memory core via the first interface circuit and for interfacing the memory core to the external circuit; anda single package housing the plurality of the first integrated circuit dies and housing the at least one second integrated circuit die.
  • 8. A memory device for use with an external circuit, the memory device comprising: a plurality of first integrated circuit dies including,a memory core comprising a plurality of memory cells, anda first interface circuit including a first interface for accessing the memory cells of the memory core;at least one second integrated circuit die. electrically coupled to the plurality of first integrated circuit dies, comprising a second interface for accessing the memory core via the first interface circuit and for interfacing the memory core to the external circuit;a first package for housing the plurality of the first integrated circuit dies; anda second package for housing the at least one second integrated circuit die.
  • 9. A memory device for use with an external circuit, the memory device comprising: at least one first integrated circuit die including,a memory core comprising a plurality of memory cells, anda first interface circuit including a first interface for accessing the memory cells of the memory core;at least one second integrated circuit die, electrically coupled to the first integrated circuit die, comprising a second interface for accessing the memory core via the first interface circuit and for interfacing the memory core to the external circuit;wherein the second interface of the at least one second integrated circuit die is further for converting protocols between the external circuit and the first interface of the first integrated circuit die.
  • 10. The memory device as set forth in claim 9, wherein: the protocols comprise,a first protocol, anda second protocol which is different from the first protocol.
  • 11. The memory device as set forth in claim 10, wherein: the first protocol comprises a synchronous protocol; andthe second protocol comprises an asynchronous protocol.
  • 12. The memory device as set forth in claim 10, wherein: the first protocol comprises a custom protocol; andthe second protocol comprises an industry standard protocol.
  • 13. The memory device as set forth in claim 1, wherein: the first interface circuit is further for providing a read operation to the memory cells.
  • 14. The memory device as set forth in claim 1, wherein: the first interface circuit is further for providing a write operation to the memory cells.
  • 15. The memory device as set forth in claim 1, wherein: the first interface circuit is further for providing activating, pre-charging and refreshing operations to the memory cells.
  • 16. The memory device as set forth in claim 1, wherein the first integrated circuit die comprises: a plurality of memory banks for partitioning the memory cells.
  • 17. The memory device as set forth in claim 16, wherein the memory banks comprise: a plurality of sub-arrays arranged in a distributed bank architecture across a plurality of physical sections such that a physical section of the memory cells comprises a plurality of sub-arrays associated with different ones of the memory banks.
  • 18. The memory device as set forth in claim 1, wherein the memory cells comprise: at least one non-volatile memory cell.
  • 19. The memory device as set forth in claim 1, wherein the memory cells comprise: at least one volatile memory cell.
  • 20. A memory device for use with an external circuit, the memory device comprising: a first integrated circuit die including,a memory core including a plurality of memory cells, anda first interface circuit having a first interface and coupled to the memory cells, for dynamically configuring an internal data rate for transferring data between the memory cells and the first interface circuit; anda second integrated circuit die, electrically coupled to the first integrated circuit die, including a second interface for accessing data from the memory core via the first interface circuit and for interfacing said memory core to the external circuit.
  • 21. The memory device as set forth in claim 20, wherein: the memory device further comprises an internal data bus for coupling data between the first integrated circuit die and the second integrated circuit die; andthe internal data bus has a configurable data width.
  • 22. The memory device as set forth in claim 20, wherein the internal data rate comprises: a configurable amount of data for pre-fetching.
  • 23. The memory device as set forth in claim 20, wherein the first integrated circuit die further comprises: at least one input for programming said data width for said internal data bus.
  • 24. The memory device as set forth in claim 20, further comprising: an external data bus, having an external data rate, for accessing data external to the memory device, wherein the internal data rate can be dynamically selected so as to be compatible with the external data rate.
  • 25. The memory device as set forth in claim 9, wherein said second interface of said second integrated circuit die for converting between a DDR2 SDRAM protocol and a DDR3 SDRAM protocol.
  • 26. The memory device as set forth in claim 9, wherein said second interface of said second integrated circuit die for converting between a first timing specification of a DDR2 SDRAM protocol and a second timing specification of a DDR2 SDRAM protocol.
  • 27. The memory device as set forth in claim 1, wherein the data rate is an internal data rate.
RELATED APPLICATIONS

This patent application claims the benefit to U.S. Provisional Patent Application entitled “Methods and Apparatus for Integrating Multi-Chip Memory Devices,” Ser. No. 60/693,631, filed on Jun. 24, 2005.

US Referenced Citations (432)
Number Name Date Kind
4525921 Carson et al. Jul 1985 A
4592019 Huang et al. May 1986 A
4646128 Carson et al. Feb 1987 A
4698748 Juzswik et al. Oct 1987 A
4706166 Go Nov 1987 A
4710903 Hereth et al. Dec 1987 A
4764846 Go Aug 1988 A
4780843 Tietjen Oct 1988 A
4841440 Yonezu et al. Jun 1989 A
4884237 Mueller et al. Nov 1989 A
4935734 Austin Jun 1990 A
4982265 Watanabe et al. Jan 1991 A
4983533 Go Jan 1991 A
5083266 Watanabe Jan 1992 A
5104820 Go et al. Apr 1992 A
5220672 Nakao et al. Jun 1993 A
5252807 Chizinsky Oct 1993 A
5257233 Schaefer Oct 1993 A
5278796 Tillinghast et al. Jan 1994 A
5282177 McLaury Jan 1994 A
5347428 Carson et al. Sep 1994 A
5384745 Konishi et al. Jan 1995 A
5388265 Volk Feb 1995 A
5432729 Carson et al. Jul 1995 A
5448511 Paurus et al. Sep 1995 A
5453434 Albaugh et al. Sep 1995 A
5467455 Gay et al. Nov 1995 A
5502667 Bertin et al. Mar 1996 A
5526320 Zagar et al. Jun 1996 A
5530836 Busch et al. Jun 1996 A
5559990 Cheng et al. Sep 1996 A
5561622 Bertin et al. Oct 1996 A
5563086 Bertin et al. Oct 1996 A
5566344 Hall et al. Oct 1996 A
5581498 Ludwig et al. Dec 1996 A
5581779 Hall et al. Dec 1996 A
5590071 Kolor et al. Dec 1996 A
5598376 Merritt et al. Jan 1997 A
5604714 Manning et al. Feb 1997 A
5606710 Hall et al. Feb 1997 A
5608262 Degani et al. Mar 1997 A
5610864 Manning Mar 1997 A
5623686 Hall et al. Apr 1997 A
5627791 Wright et al. May 1997 A
5640337 Huang et al. Jun 1997 A
5640364 Merritt et al. Jun 1997 A
5652724 Manning Jul 1997 A
5661677 Rondeau et al. Aug 1997 A
5661695 Zagar et al. Aug 1997 A
5668773 Zagar et al. Sep 1997 A
5675549 Ong et al. Oct 1997 A
5680342 Frankeny Oct 1997 A
5682354 Manning Oct 1997 A
5692202 Kardach et al. Nov 1997 A
5696732 Zagar et al. Dec 1997 A
5702984 Bertin et al. Dec 1997 A
5703813 Manning et al. Dec 1997 A
5706247 Merritt et al. Jan 1998 A
5717654 Manning Feb 1998 A
5721859 Manning Feb 1998 A
5724288 Cloud et al. Mar 1998 A
5729503 Manning Mar 1998 A
5748914 Barth et al. May 1998 A
5752045 Chen May 1998 A
5757703 Merritt et al. May 1998 A
5798961 Heyden et al. Aug 1998 A
5802010 Zagar et al. Sep 1998 A
5812488 Zagar et al. Sep 1998 A
5831931 Manning Nov 1998 A
5831932 Merritt et al. Nov 1998 A
5835435 Bogin et al. Nov 1998 A
5838165 Chatter Nov 1998 A
5838177 Keeth Nov 1998 A
5843807 Burns Dec 1998 A
5845108 Yoo et al. Dec 1998 A
5850368 Ong et al. Dec 1998 A
5859792 Rondeau et al. Jan 1999 A
5860106 Domen et al. Jan 1999 A
5870347 Keeth et al. Feb 1999 A
5875142 Chevallier Feb 1999 A
5884088 Kardach et al. Mar 1999 A
5901105 Ong et al. May 1999 A
5903500 Tsang et al. May 1999 A
5905688 Park May 1999 A
5907512 Parkinson et al. May 1999 A
5915105 Farmwald et al. Jun 1999 A
5917758 Keeth Jun 1999 A
5923611 Ryan Jul 1999 A
5926452 Park et al. Jul 1999 A
5943254 Bakeman, Jr. et al. Aug 1999 A
5946265 Cowles Aug 1999 A
5949254 Keeth Sep 1999 A
5953215 Karabatsos Sep 1999 A
5953263 Farmwald et al. Sep 1999 A
5954804 Farmwald et al. Sep 1999 A
5956233 Yew et al. Sep 1999 A
5962435 Mao et al. Oct 1999 A
5963463 Rondeau et al. Oct 1999 A
5963464 Dell et al. Oct 1999 A
5963504 Manning Oct 1999 A
5966724 Ryan Oct 1999 A
5973392 Senba et al. Oct 1999 A
5995443 Farmwald et al. Nov 1999 A
6001671 Fjelstad Dec 1999 A
6002613 Cloud et al. Dec 1999 A
6002627 Chevallier Dec 1999 A
6014339 Kobayashi et al. Jan 2000 A
6016282 Keeth Jan 2000 A
6032214 Farmwald et al. Feb 2000 A
6032215 Farmwald et al. Feb 2000 A
6034916 Lee Mar 2000 A
6034918 Farmwald et al. Mar 2000 A
6035365 Farmwald et al. Mar 2000 A
6038195 Farmwald et al. Mar 2000 A
6038673 Benn et al. Mar 2000 A
6044032 Li Mar 2000 A
6069504 Keeth May 2000 A
6070217 Connolly et al. May 2000 A
6073223 McAllister et al. Jun 2000 A
6075730 Barth et al. Jun 2000 A
6078546 Lee Jun 2000 A
6079025 Fung Jun 2000 A
6084434 Keeth Jul 2000 A
6088290 Ohtake et al. Jul 2000 A
RE36839 Simmons et al. Aug 2000 E
6101152 Farmwald et al. Aug 2000 A
6101612 Jeddeloh Aug 2000 A
6108795 Jeddeloh Aug 2000 A
6154370 Degani et al. Nov 2000 A
6166991 Phelan Dec 2000 A
6182184 Farmwald et al. Jan 2001 B1
6208168 Rhee Mar 2001 B1
6222739 Bhakta et al. Apr 2001 B1
6233192 Tanaka May 2001 B1
6240048 Matsubara May 2001 B1
6243282 Rondeau et al. Jun 2001 B1
6252807 Suzuki et al. Jun 2001 B1
6260097 Farmwald et al. Jul 2001 B1
6260154 Jeddeloh Jul 2001 B1
6262938 Lee et al. Jul 2001 B1
6266285 Farmwald et al. Jul 2001 B1
6295572 Wu Sep 2001 B1
6298426 Ajanovic Oct 2001 B1
6314051 Farmwald et al. Nov 2001 B1
6324120 Farmwald et al. Nov 2001 B2
6326810 Keeth Dec 2001 B1
6327664 Dell et al. Dec 2001 B1
6330683 Jeddeloh Dec 2001 B1
6338108 Motomura Jan 2002 B1
6338113 Kubo et al. Jan 2002 B1
6353561 Funyu et al. Mar 2002 B1
6356105 Volk Mar 2002 B1
6356500 Cloud et al. Mar 2002 B1
6362656 Rhee Mar 2002 B2
6363031 Phelan Mar 2002 B2
6378020 Farmwald et al. Apr 2002 B2
6381188 Choi et al. Apr 2002 B1
6381668 Lunteren Apr 2002 B1
6392304 Butler May 2002 B1
6418034 Weber et al. Jul 2002 B1
6426916 Farmwald et al. Jul 2002 B2
6430103 Nakayama et al. Aug 2002 B2
6437600 Keeth Aug 2002 B1
6438057 Ruckerbauer Aug 2002 B1
6442698 Nizar Aug 2002 B2
6452826 Kim et al. Sep 2002 B1
6452863 Farmwald et al. Sep 2002 B2
6453400 Maesako et al. Sep 2002 B1
6457095 Volk Sep 2002 B1
6459651 Lee et al. Oct 2002 B1
6487102 Halbert et al. Nov 2002 B1
6489669 Shimada et al. Dec 2002 B2
6493789 Ware et al. Dec 2002 B2
6496440 Manning Dec 2002 B2
6496897 Ware et al. Dec 2002 B2
6510097 Fukuyama Jan 2003 B2
6510503 Gillingham et al. Jan 2003 B2
6521984 Matsuura Feb 2003 B2
6526471 Shimomura et al. Feb 2003 B1
6526484 Stacovsky et al. Feb 2003 B1
6545895 Li et al. Apr 2003 B1
6546446 Farmwald et al. Apr 2003 B2
6553450 Dodd et al. Apr 2003 B1
6560158 Choi et al. May 2003 B2
6563337 Dour May 2003 B2
6563759 Yahata et al. May 2003 B2
6564281 Farmwald et al. May 2003 B2
6574150 Suyama et al. Jun 2003 B2
6584037 Farmwald et al. Jun 2003 B2
6587912 Leddige et al. Jul 2003 B2
6590822 Hwang et al. Jul 2003 B2
6594770 Sato et al. Jul 2003 B1
6597617 Ooishi et al. Jul 2003 B2
6614700 Dietrich et al. Sep 2003 B2
6618791 Dodd et al. Sep 2003 B1
6621760 Ahmad et al. Sep 2003 B1
6630729 Huang Oct 2003 B2
6631086 Bill et al. Oct 2003 B1
6639820 Khandekar et al. Oct 2003 B1
6646939 Kwak Nov 2003 B2
6650594 Lee et al. Nov 2003 B1
6657634 Sinclair et al. Dec 2003 B1
6657918 Foss et al. Dec 2003 B2
6657919 Foss et al. Dec 2003 B2
6658530 Robertson et al. Dec 2003 B1
6665224 Lehmann et al. Dec 2003 B1
6665227 Fetzer Dec 2003 B2
6683372 Wong et al. Jan 2004 B1
6697295 Farmwald et al. Feb 2004 B2
6701446 Tsern et al. Mar 2004 B2
6705877 Li et al. Mar 2004 B1
6708144 Merryman et al. Mar 2004 B1
6714891 Dendinger Mar 2004 B2
6724684 Kim Apr 2004 B2
6744687 Koo et al. Jun 2004 B2
6747887 Halbert et al. Jun 2004 B2
6751113 Bhakta et al. Jun 2004 B2
6751696 Farmwald et al. Jun 2004 B2
6754129 Khatri et al. Jun 2004 B2
6754132 Kyung Jun 2004 B2
6757751 Gene Jun 2004 B1
6762948 Kyun et al. Jul 2004 B2
6765812 Anderson Jul 2004 B2
6771526 LaBerge Aug 2004 B2
6772359 Kwak et al. Aug 2004 B2
6779097 Gillingham et al. Aug 2004 B2
6795899 Dodd et al. Sep 2004 B2
6799241 Kahn et al. Sep 2004 B2
6807598 Farmwald et al. Oct 2004 B2
6816991 Sanghani Nov 2004 B2
6819602 Seo et al. Nov 2004 B2
6819617 Hwang et al. Nov 2004 B2
6820163 McCall et al. Nov 2004 B1
6820169 Wilcox et al. Nov 2004 B2
6826104 Kawaguchi et al. Nov 2004 B2
6839290 Ahmad et al. Jan 2005 B2
6845055 Koga et al. Jan 2005 B1
6847582 Pan Jan 2005 B2
6850449 Takahashi Feb 2005 B2
6862202 Schaefer Mar 2005 B2
6862249 Kyung Mar 2005 B2
6862653 Dodd et al. Mar 2005 B1
6873534 Bhakta et al. Mar 2005 B2
6878570 Lyu et al. Apr 2005 B2
6894933 Kuzmenka et al. May 2005 B2
6898683 Nakamura May 2005 B2
6922371 Takahashi et al. Jul 2005 B2
6930900 Bhakta et al. Aug 2005 B2
6930903 Bhakta et al. Aug 2005 B2
6938119 Kohn et al. Aug 2005 B2
6943450 Fee et al. Sep 2005 B2
6951982 Chye et al. Oct 2005 B2
6961281 Wong et al. Nov 2005 B2
6968416 Moy Nov 2005 B2
6968419 Holman Nov 2005 B1
6970968 Holman Nov 2005 B1
6986118 Dickman Jan 2006 B2
6992501 Rapport Jan 2006 B2
6992950 Foss et al. Jan 2006 B2
7000062 Perego et al. Feb 2006 B2
7003618 Perego et al. Feb 2006 B2
7003639 Tsern et al. Feb 2006 B2
7007175 Chang et al. Feb 2006 B2
7010642 Perego et al. Mar 2006 B2
7010736 Teh et al. Mar 2006 B1
7026708 Cady et al. Apr 2006 B2
7028234 Huckaby et al. Apr 2006 B2
7033861 Partridge et al. Apr 2006 B1
7035150 Streif et al. Apr 2006 B2
7045396 Crowley et al. May 2006 B2
7045901 Lin et al. May 2006 B2
7046538 Kinsley et al. May 2006 B2
7053470 Sellers et al. May 2006 B1
7053478 Roper et al. May 2006 B2
7061784 Jakobs et al. Jun 2006 B2
7061823 Faue et al. Jun 2006 B2
7066741 Burns et al. Jun 2006 B2
7079441 Partsch et al. Jul 2006 B1
7079446 Murtagh et al. Jul 2006 B2
7085941 Li Aug 2006 B2
7089438 Raad Aug 2006 B2
7103730 Saxena et al. Sep 2006 B2
7126399 Lee Oct 2006 B1
7149824 Johnson Dec 2006 B2
7205789 Karabatsos Apr 2007 B1
7228264 Barrenscheen et al. Jun 2007 B2
7245541 Janzen Jul 2007 B2
7269708 Ware Sep 2007 B2
7299330 Gillingham et al. Nov 2007 B2
7379316 Rajan May 2008 B2
7386656 Rajan et al. Jun 2008 B2
7392338 Rajan et al. Jun 2008 B2
20010000822 Dell et al. May 2001 A1
20010021106 Weber et al. Sep 2001 A1
20010021137 Kai et al. Sep 2001 A1
20020019961 Blodgett Feb 2002 A1
20020034068 Weber et al. Mar 2002 A1
20020038405 Leddige et al. Mar 2002 A1
20020041507 Woo et al. Apr 2002 A1
20020064073 Chien May 2002 A1
20020064083 Ryu et al. May 2002 A1
20020089831 Forthun Jul 2002 A1
20020124195 Nizar Sep 2002 A1
20020145900 Schaefer Oct 2002 A1
20020174274 Wu et al. Nov 2002 A1
20020184438 Usui Dec 2002 A1
20030002262 Benisek et al. Jan 2003 A1
20030021175 Kwak Jan 2003 A1
20030035312 Halbert et al. Feb 2003 A1
20030039158 Horiguchi et al. Feb 2003 A1
20030061458 Wilcox et al. Mar 2003 A1
20030093614 Kohn et al. May 2003 A1
20030101392 Lee May 2003 A1
20030105932 David et al. Jun 2003 A1
20030117875 Lee et al. Jun 2003 A1
20030126338 Dodd et al. Jul 2003 A1
20030131160 Hampel et al. Jul 2003 A1
20030158995 Lee et al. Aug 2003 A1
20030164539 Yau Sep 2003 A1
20030182513 Dodd et al. Sep 2003 A1
20030183934 Barrett Oct 2003 A1
20030189868 Riesenman et al. Oct 2003 A1
20030189870 Wilcox Oct 2003 A1
20030191888 Klein Oct 2003 A1
20030191915 Saxena et al. Oct 2003 A1
20030200382 Wells et al. Oct 2003 A1
20030200474 Li Oct 2003 A1
20030205802 Segaram Nov 2003 A1
20030206476 Joo Nov 2003 A1
20030217303 Chua-Eoan et al. Nov 2003 A1
20030227798 Pax Dec 2003 A1
20030229821 Ma Dec 2003 A1
20030231540 Lazar et al. Dec 2003 A1
20030231542 Zaharinova-Papazova et al. Dec 2003 A1
20040027902 Ooishi et al. Feb 2004 A1
20040034732 Valin et al. Feb 2004 A1
20040044808 Salmon et al. Mar 2004 A1
20040047228 Chen Mar 2004 A1
20040057317 Schaefer Mar 2004 A1
20040064767 Huckaby et al. Apr 2004 A1
20040088475 Streif et al. May 2004 A1
20040117723 Foss Jun 2004 A1
20040125635 Kuzmenka Jul 2004 A1
20040133736 Kyung Jul 2004 A1
20040139359 Samson et al. Jul 2004 A1
20040145963 Byon Jul 2004 A1
20040151038 Ruckerbauer et al. Aug 2004 A1
20040174765 Seo et al. Sep 2004 A1
20040178824 Pan Sep 2004 A1
20040184324 Pax Sep 2004 A1
20040186956 Perego et al. Sep 2004 A1
20040188704 Halbert et al. Sep 2004 A1
20040196732 Lee Oct 2004 A1
20040228166 Braun et al. Nov 2004 A1
20040228203 Koo Nov 2004 A1
20040230932 Dickmann Nov 2004 A1
20040256638 Perego et al. Dec 2004 A1
20040257847 Matsui et al. Dec 2004 A1
20040260957 Jeddeloh et al. Dec 2004 A1
20040264255 Royer Dec 2004 A1
20040268161 Ross Dec 2004 A1
20050018495 Bhakta et al. Jan 2005 A1
20050021874 Georgiou et al. Jan 2005 A1
20050024963 Jakobs et al. Feb 2005 A1
20050028038 Pomaranski et al. Feb 2005 A1
20050036350 So et al. Feb 2005 A1
20050041504 Perego et al. Feb 2005 A1
20050044303 Perego et al. Feb 2005 A1
20050044305 Jakobs et al. Feb 2005 A1
20050047192 Matsui et al. Mar 2005 A1
20050071543 Ellis et al. Mar 2005 A1
20050078532 Ruckerbauer et al. Apr 2005 A1
20050081085 Ellis et al. Apr 2005 A1
20050099834 Funaba et al. May 2005 A1
20050102590 Norris et al. May 2005 A1
20050105318 Funaba et al. May 2005 A1
20050138267 Bains et al. Jun 2005 A1
20050139977 Nishio et al. Jun 2005 A1
20050141199 Chiou et al. Jun 2005 A1
20050149662 Perego Jul 2005 A1
20050152212 Yang et al. Jul 2005 A1
20050156934 Perego et al. Jul 2005 A1
20050166026 Ware et al. Jul 2005 A1
20050193163 Perego Sep 2005 A1
20050194991 Dour et al. Sep 2005 A1
20050195629 Leddige et al. Sep 2005 A1
20050204111 Natarajan Sep 2005 A1
20050207255 Perego et al. Sep 2005 A1
20050210196 Perego et al. Sep 2005 A1
20050223179 Perego et al. Oct 2005 A1
20050224948 Lee et al. Oct 2005 A1
20050235131 Ware Oct 2005 A1
20050249011 Maeda Nov 2005 A1
20050259504 Murtugh et al. Nov 2005 A1
20050265506 Foss et al. Dec 2005 A1
20050269715 Yoo Dec 2005 A1
20050278474 Perersen et al. Dec 2005 A1
20050281123 Bell et al. Dec 2005 A1
20050283572 Ishihara Dec 2005 A1
20050285174 Saito et al. Dec 2005 A1
20050289292 Morrow et al. Dec 2005 A1
20060010339 Klein Jan 2006 A1
20060039205 Cornelius Feb 2006 A1
20060041711 Miura et al. Feb 2006 A1
20060041730 Larson Feb 2006 A1
20060044913 Klein et al. Mar 2006 A1
20060049502 Goodwin et al. Mar 2006 A1
20060050574 Streif et al. Mar 2006 A1
20060067141 Perego et al. Mar 2006 A1
20060085616 Zeighami et al. Apr 2006 A1
20060090054 Choi et al. Apr 2006 A1
20060112214 Yeh May 2006 A1
20060117160 Jackson et al. Jun 2006 A1
20060118933 Haba Jun 2006 A1
20060120193 Casper Jun 2006 A1
20060123265 Ruckerbauer et al. Jun 2006 A1
20060126369 Raghuram Jun 2006 A1
20060129712 Raghuram Jun 2006 A1
20060129755 Raghuram Jun 2006 A1
20060133173 Jain et al. Jun 2006 A1
20060149982 Vogt Jul 2006 A1
20060176744 Stave Aug 2006 A1
20060179333 Brittain et al. Aug 2006 A1
20060179334 Brittain et al. Aug 2006 A1
20060198178 Kinsley et al. Sep 2006 A1
20060203590 Mori et al. Sep 2006 A1
20060206738 Jeddeloh et al. Sep 2006 A1
20060233012 Sekiguchi et al. Oct 2006 A1
20060262586 Solomon et al. Nov 2006 A1
20070070669 Tsern et al. Mar 2007 A1
20070088995 Tsern et al. Apr 2007 A1
20080065820 Gillingham et al. Mar 2008 A1
Foreign Referenced Citations (17)
Number Date Country
102004053316 May 2006 DE
102005036528 Feb 2007 DE
2441726 Mar 2008 GB
01171047 Jul 1989 JP
03-029357 Feb 1991 JP
03276487 Dec 1991 JP
03286234 Dec 1991 JP
07-141870 Jun 1995 JP
08077097 Mar 1996 JP
08077097 Mar 1996 JP
11-149775 Jun 1999 JP
2002025255 Jan 2002 JP
1020040062717 Jul 2004 KR
WO9505676 Feb 1995 WO
WO9900734 Jan 1999 WO
WO2007002324 Jan 2007 WO
WO 2007038225 Apr 2007 WO
Related Publications (1)
Number Date Country
20070050530 A1 Mar 2007 US
Provisional Applications (1)
Number Date Country
60693631 Jun 2005 US