Random access memory architecture and serial interface with continuous packet handling capability

[0001] As the need for bandwidth expands in information and communication networks, bottlenecks develop in the switching hardware due to increases in the number of ports and throughput per port in routing. To handle the increased bandwidth, architectures arose that rely upon increased internal bus widths, increased bandwidths on the busses, and decentralized processing. A block diagram of an example of the architecture of a known router system 100 is shown in FIG. 1. Incoming packets can be temporarily stored in a local buffer 102 until one or more network layer processors, shown as application specific integrated circuits (ASICs) 104 and 106, classifies and determines the forwarding address of the packet. The packets are then routed to the target destination.

[0002] Buffer 102 includes gigabytes of SRAM arranged in a very wide data bus (512 bits or more) 108 to allow several simultaneous search operations. Router system 100 analyzes the headers of incoming packets for the appropriate types of operations to be performed on the packet. In the case of TCP/IP route look-up, for example, router system 100 matches a combination of source/destination addresses with entries in a routing table for an exact match or a longest prefix match. Content addressable memory (CAM) 110 offloads ASIC 104 by processing a limited set of lookup instructions. In other applications such as load balancing and URL routing, router system 100 performs more complex search rules and routing data structures.

[0003] A successful search will result in router system 100 modifying the header and optionally, the payload, with new information. A packet header may be modified for its output port/queue, output connection, or additional labels for switching, for example. The packets are then queued by priority as well as latency to absorb the time needed to implement the search/modify steps. Output queuing reorders packets in the event multiple packets are searched in parallel.

[0004] In the example shown in FIG. 1, data bus 108 is a 640 bit bus with a speed of 133 MHz or 167 MHz. Greater efficiency can be achieved with large data packets, while transfers over data bus 108 are very inefficient for small data packets. ASICS 104 and 106 have a pin count of over 1000 pins. The number of pins cannot be expanded without changing ASICs 104, 106. The bandwidth of router system 100 is therefore limited.

[0005] If an ASIC with an expanded number of pins is used, other problems arise which increase production costs and limit scalability of router system 100. For example, the memory interface presents board layout problems, including positioning components to avoid long data bus lines; routing wide data bus lines to memory and processor devices with a large number of pins; and multidrop connections on all busses. Other problems include limited bandwidth per pin, and limited memory access time. Additionally, address fan-out problems arise with increasing address bandwidth and address loading and buffering. Further, noise issues arise with heavy bus loading and heavy bus drivers, line termination, bus contention, and the length of the data bus lines. There are also logistical problems with bus arbitration, transferring small amounts of data on very wide busses, bottlenecks at the memory interface, and bus/clock skew and bit alignment. Numerous memory devices switching on many pins at very high speed also dissipates a large amount of power. Additionally, because increasing bus widths, bus bandwidths, bus loading, ASIC size, and memory depth brings diminishing returns, router system 100 cannot efficiently be expanded to handle greater packet bandwidth.

[0006] It is therefore desirable to provide a router that is cost effective and scalable, and alleviates many of the problems listed above.

[0007] According to an embodiment of the invention, an apparatus may include a first processing unit operable to receive serial data input, a switch operable to receive data to be transmitted, a first random access memory device coupled to the first processing unit via a write input port, wherein the first processing unit is operable to transmit data packets to the first random access memory device via the write input port, and a second random access memory device coupled to the first random access memory device, the second random access memory device being further coupled to transmit the data to be transmitted to the switch. The first and second random access memory devices may be configured as described in the embodiments below.

[0008] There may be a plurality of random access memory devices coupled between the first random access memory device and the second random access memory device, where each of the plurality of random access memory devices may be configured according to one of the embodiments described below. Each of the plurality of random access memory devices may be coupled in series to another of the plurality of random access memory devices. A write output port of each of the plurality of random access memory devices may be coupled to a write input port of the next of the series of the plurality of random access memory devices. A read output port of each of the plurality of random access memory devices may be coupled to a read input port of the next of the series of the plurality of random access memory devices. In some embodiments, each of the random access memory devices is operable to generate an overflow signal to indicate to the next random access memory device in the series that the data to be written or read is a continuation from the previous random access memory device. Each random access memory device may be operable to generate a command signal to indicate to the next random access memory device in the series that the continuation of data is to be written or read.

[0009] According to an embodiment of the invention, a random access memory device comprises a random access memory module, a write input port coupled to receive packets that include data and command information, a decoder coupled to receive the packets from the write input port, and to access the random access memory module to perform the commands in the packets, a read input port operable to receive bypass data, and a read output port coupled to output the bypass data and data read from the random access memory module.

[0010] The random access memory device may also include a write output port coupled to receive the packets from the write input port and to transmit the packets to another random access memory device. The random access memory device may also comprise a buffer coupled between the write input port and the write output port. The random access memory device may include a decoder between the write input port and the write output port, and an encoder coupled between the decoder and the write output port. The decoder may include a phase lock loop.

[0011] The random access memory device may also include a first multiplexer coupled between the read input port and the read output port, operable to output an idle signal when the read input port is not transmitting data to the read output port. The random access memory device may include a second multiplexer, coupled between the random access memory module and the read output port, where the second multiplexer is coupled to receive the output of the first multiplexer and to output the output of the first multiplexer when read data is not being transmitted from the random access memory module. The random access memory device may also include an encoder coupled between the random access module and the second multiplexer. The random access memory device may include a buffer coupled between the second multiplexer and the read output port. The random access memory device may also include a second decoder coupled between the read input port and the first multiplexer, wherein the second decoder includes a delay lock loop.

[0012] In some embodiments, the random access memory devices are operable to decode data packets and determine whether a data packet includes at least one of: a start of frame field, a control word, an address field, a data field, and an idle indicator. The control word may include at least one of: a command to be performed, and the random access memory device to perform the command.

[0013] According to an embodiment of the invention, a method for routing data includes receiving serial data input in a first processing unit, generating a data packet based on the serial data input, transmitting the data packet to a first random access memory device via a write input port, decoding the data packet, determining whether to perform a command in the first random access memory device based on information in the data packet, and transmitting the data packet to a second random access memory device, wherein a write input port of the second random access memory device is coupled in series to a write output port of the first random access memory device. The method may further include determining whether to perform a command in the second random access memory device based on the information in the data packet. The method may further include coupling a plurality of random access memory devices in series between the first random access memory device and the second random access memory device, wherein each of the plurality of random access memory devices includes a write input port, a write output port, a read input port, and a read output port, transmitting the data packet from the first random access memory device to each random access memory device, wherein the write output port of each random access memory device is coupled in series to the write input port of the next random access memory device, and determining whether to perform a command in each random access memory device based on the information in the data packet.

[0014] The method may further include generating an overflow signal in one of the plurality of random access memory devices to indicate to the next random access memory device that a write command or read command is being continued from the one of the plurality of random access memory devices. The method may further include terminating performance of a first write command when a second write command is received, when a read command is received, or when a stop write command is received.

[0015] The present invention may be better understood, and its numerous objects, features, and advantages made even more apparent to those skilled in the art by referencing the detailed description and accompanying drawings of the embodiments described below.

[0016] FIG. 1 is a block diagram of an example of components included in a high speed router known in the art.

[0017] FIG. 2 is a block diagram of components included in a router according to an embodiment of the present invention.

[0018] FIG. 3a is a block diagram of an example of a SRAM device that can be utilized in the router of FIG. 2.

[0019] FIG. 3b is a block diagram of an alternate SRAM device that can be utilized in the router of FIG. 2.

[0020] FIG. 4 is a block diagram showing an embodiment of a configuration of the bank of SRAM devices of FIG. 2.

[0021] FIG. 5 shows an example of an address and data format for two packets that can be processed by the SRAM devices of FIG. 3a or 3b.

[0022] FIG. 6 shows an example of a packet write that can be processed by the SRAM devices of FIG. 3a or 3b.

[0023] FIG. 7 shows an example of a packet write interrupted by a packet read that can be processed by the SRAM devices of FIG. 3a or 3b.

[0024] FIG. 8 shows an example of a read data that can be output on the read output port of the SRAM devices of FIG. 3a or 3b in response to two read requests.

[0025] FIG. 9 shows an example of a packet write followed immediately by a packet read of the data that can be processed by the SRAM devices of FIG. 3a or 3b.

[0026] FIG. 10a shows an example of a read request for two minimum sized packets requested through the write input port.

[0027] FIG. 10b shows the output on the read port resulting from the read commands received in the packets of FIG. 10a.

[0028] FIG. 11a shows an example of input and output bandwidths that can be achieved with a known high speed router.

[0029] FIG. 11b shows an example of input and output bandwidth that can be achieved by a high speed router configured with SRAM devices of FIG. 3a or 3b.

[0030] A configuration of static random access memory (SRAM) devices in accordance with an embodiment of the present invention replaces the data bus 108 (FIG. 1) with high speed, point-to-point, serial data connections. Accordingly, the pin count on the controlling ASIC 104 (FIG. 1) can be significantly reduced while simultaneously achieving at least an order of magnitude increase in total bandwidth. Further, increasing the memory capacity of SRAM 102 (FIG. 1) adds bus loading on bus 108 and possibly extra pins on ASIC 104. Further, increasing the size and number of memory storage devices in SRAM 102 creates difficulties in routing bus 108 and address distribution. SRAM devices configured in accordance with an embodiment of the present invention allow an increase in the amount of memory without requiring additional pins on an ASIC and can be distributed serially in chains that are connected to the ASIC at both ends. Such a configuration simplifies real estate problems in the vicinity of the ASIC and alleviates problems associated with routing additional bus lines to ASIC 104. Further, the bandwidth per pin of an ASIC configured in accordance with embodiments of the present invention can be significantly increased, and memory reads and writes can occur simultaneously and continuously as further described hereinbelow.

[0031] Referring now to FIG. 2, a block diagram of components included in a router 200 according to an embodiment of the present invention is shown. Router 200 includes an ASIC 202, a CAM 110, SRAM devices 204 associated with CAM 110, a bank of SRAM devices 206, and a switch 208. Each physical input has a direct connection to ASIC 202. ASIC 202 transmits the header from each packet to CAM 110 for translation, and stores the rest of the packet, also referred to as the payload, in the bank of SRAM devices 206. The payload remains in the bank of SRAM devices 206 until CAM 110 returns the header with the new routing information for the packet to ASIC 202. ASIC 202 retrieves the payload, combines the new header with the corresponding payload, and transmits the packet to switch 208 via the bank of SRAM devices 206.

[0032] FIG. 3a is a block diagram of an embodiment of an SRAM device 300 that can be utilized in the bank of SRAM devices 206 of FIG. 2. In the embodiment shown, data is transmitted directly from a write input port 302 to a write output port 304 under all conditions. Control data, write data, and read requests are input to SRAM device 300 through write input port 302. When a valid write command is in progress, the packet is transmitted to a decoder and phase lock loop (PLL) 306. The data and the address where the data will be stored in an SRAM module 308 is decoded from the packet and transmitted to SRAM module 308, which stores the data.

[0033] When a read command is received in write input port 302, the data is retrieved from SRAM module 308 and transmitted to an encoder 310, a buffer 312, and read output port 314. When a read is in progress, a multiplexer 316 suspends transmission of bypass data through read input port 318 until the requested data from SRAM module 308 has been sent. When data is not being output on read output port 314, then bypass data is transmitted directly to read output port 314 via multiplexer 316 and buffer 312. If a read command is not in progress, and no bypass data is being transmitted (i.e., read input port 318 is not toggling), then an IDLE signal 320 is transmitted to read output port 314 via multiplexer 316 and buffer 312.

[0034] FIG. 3b is a block diagram of another embodiment of an SRAM device 330 that can be utilized in the bank of SRAM devices 206 of FIG. 2. In the embodiment of SRAM device 330 shown in FIG. 3b, data is transmitted from write input port 302 to decoder and PLL 306, encoder 310, and then to write output port 304 under all conditions. Decoder and PLL 306 generates clock signals for a decoder and delay lock loop (DLL) 332 to correct the phase of the bypass data. The clock signals from decoder and PLL 306 are also used to clock the read output port 314 and the write output port 304, so that data transmitted from write output port 304 is synchronized with data transmitted from read output port 314. The timing is set up to be equal in both read and write paths of SRAM device 330, so that individual devices in a bank of SRAM devices (such as bank of SRAM devices 206 of FIG. 2) can insert data into the read data stream without interfering with data from other devices. When a valid write command is in progress, the data and the address where the data will be stored in SRAM module 308 is transmitted to SRAM module 308, which stores the data.

[0035] When a read command is received in write input port 302, the data is retrieved from SRAM module 308 and transmitted to an encoder 334 and read output port 314. When a read is in progress, multiplexer 316 suspends transmission of bypass data through read input port 318 until the requested data from SRAM module 308 has been sent. When data is not being output on read output port 314, then bypass data is transmitted to read output port 314 via decoder and DLL 332 and encoder 334. If no read command is in progress, and there is no bypass data being transmitted, then an IDLE signal 320 is transmitted to read output port 314 via decoder and DLL 332 and encoder 334.

[0036] FIG. 4 shows a block diagram of an embodiment of system memory architecture for one or more SRAM series 400, 401 that can be utilized in the bank of SRAM devices 206 of FIG. 2. Each SRAM series 400, 401 includes one write input port 402 to receive input from ASIC 202 and one read output port 404 to output read data to switch 208. Note that ASIC 202 and switch 208 are shown in the same component in FIG. 4 to illustrate the architecture of the SRAM series 400, 401. Additionally, multiple SRAM devices 300 (FIG. 3a) and SRAM devices 330 (FIG. 3b) can be utilized in SRAM series 400, 401, with SRAM devices 300 being shown as one possible configuration.

[0037] In the architecture shown, SRAM series 400 is dedicated to receiving and storing data from a serial input 406 and supplying the data to switch 208. Similarly, SRAM series 401 is dedicated to receiving and storing data from a serial input 408 and supplying the data to switch 208. Additional SRAM devices 300 can be added to a SRAM series 400, 401 to increase storage capacity without increasing the number of pins on ASIC 202. Further, additional SRAM series 400, 401 can be added to handle a corresponding number of additional serial inputs.

[0038] The read output ports 314 and the write output ports 304 of each SRAM device 300 are connected to the read input port 318 and write input port 302, respectively, of the next SRAM device 300 in the SRAM series 400, 401. Notably, SRAM devices 300 can be added or removed from the SRAM series 400, 401 to meet latency requirements. For example, in one embodiment, a series of fifty-two (52) eighteen (18) Megabit SRAM devices 300 is required to reach a latency of 300 milliseconds.

[0039] In some embodiments, SRAM devices 300 and 330 are configured to manage overflow data. If the amount of data to be written is larger than the amount of memory space available in the SRAM device 300 specified in the CNTL word, the data overflows to the next SRAM 300 in the SRAM series 400, 401. In this manner, the data from a packet is stored in a contiguous block that can span portions of one or more SRAM devices 300. When a read operation, such as the read operation illustrated in FIG. 8, would access the last data word in one of the SRAM devices 300 but the read operation is not halted, an overflow occurs. An overflow signal on a connection such as connection 410 can indicate to the next SRAM device 300 or 330 in the SRAM series 400, 401 that the read operation should continue in the subsequent device starting at the lowest address. In SRAM devices 300 or 330, connection 410 (as an output) indicates that the address at the SRAM is at terminal count. Connection 410 (as an input) indicates that the SRAM address should be set to zero and incremented from there.

[0040] Similarly, an overflow signal on a connection such as connection 412 can indicate that an overflow occurs during a write operation. When a write operation, such as the write operation illustrated in FIG. 6, would write to the last data word in an SRAM device 300, but the write operation is not halted, the overflow signal on connection 412 would indicate to the next SRAM device 300 in the SRAM series 400, 401 that the write operation should continue in the subsequent device starting at the lowest address. In SRAM device 300 or 330, connection 412 (as an output) would be an indication that the address at the SRAM is at terminal count. Connection 412 (as an input) would indicate that the SRAM address should be set to zero and incremented from there. This method of managing overflow is referred to as cascading first-in-first-out (FIFO). Notably, the overflow signal on connection 410 or connection 412 also indicates that the data should be read from or written to the beginning of the memory in the SRAM device 300, thereby alleviating the need for ASIC 202 to supply a new address or device selection to the next SRAM device 300.

[0041] FIG. 5 is an example of an address and data format for a packet 500 that can be processed by the SRAM devices 300, 330 of FIG. 3a or 3b. When packet 500 is input to the write input port 302, the start of frame (SOF) parameter initiates a read or write transaction by apportioning the data in one or more frames and synchronizing the serial bit stream. In one embodiment, the control (CNTL) word includes 8 bits that can be set to one of several read or write modes. For example, two bits can be set to command the following read or write modes:

	00	READ until stop command
	01	WRITE until stop command
	10	STOP command (remaining 6 bits in the CNTL
		word are set to indicate:
		000000	Stop Write
		100000	Stop Read)
	11	READ UNTIL STOP WITH DATA TAGGING command

[0042] Note that an address is not required for the STOP command. For the READ and WRITE commands, the remaining 6 bits in the CNTL word can also be used to select a SRAM device 300 (FIG. 3). This allows up to 64 SRAM devices 300 to be cascaded. A greater or lesser number of bits can be included in the CNTL word to accommodate any number of SRAM devices 300, as required by a particular implementation.

[0043] In the embodiment shown, the address (ADD) field includes 24 bits that are set to indicate the starting location of the data to be read or written when a READ or WRITE command is input. A minimum amount of memory space, such as 1 byte, can be allocated to each address. Other amounts of minimum memory space and number of bits in the address field can be implemented.

[0044] If a READ UNTIL STOP WITH DATA TAGGING command is input, a number of bits in each DATA field can be used to store the data tag. For example, in one implementation, 16 bits of the DATA field are used to store tag information, such as a sequence number identifying the data to subsequent devices.

[0045] When the WRITE command is received, data is written into one or more of the SRAM devices 300, 330 (FIGS. 3a and 3b) until the end of the data is reached. The WRITE command can be suspended when the STOP WRITE command is detected.

[0046] FIG. 6 shows an example of a packet write that can be input to write input port 302 and processed by SRAM devices 300, 330 of FIG. 3a or 3b. Packet write 600 includes the SOF field to indicate the start of a new control word and initiate the transaction by framing the data and synchronizing the serial bit stream. The CNTL word indicates that the WRITE command is to be performed. A portion of the CNTL word can indicate the SRAM device 300 in the SRAM series 400, 401 (FIG. 4) into which the data will be written. The ADDR fields indicate the starting address in the SRAM device 300 in the SRAM series 400, 401 (FIG. 4) for writing the data. In one embodiment, each address requires a minimum of 32 bytes of address space. A variety of different size packets can be used within the same SRAM series 400, 401.

[0047] In some embodiments, the STOP WRITE command is performed when the command is detected in the CNTL word after the SOF field. A STOP WRITE can also be initiated by starting a new WRITE command. In such an embodiment, the WRITE command terminates all existing writes in all SRAM devices 300 in the SRAM series 400, 401 (FIG. 4). The SRAM device 300 specified in the CNTL word starts a new write transaction for the write data in the packet write 600.

[0048] According to some embodiments, both read and write operations may be performed in a bank of SRAM devices (such as bank of SRAM devices 206 of FIG. 2) at the same time. Referring now to FIG. 7, an example of a packet write 702 interrupted by a packet read 704 that can be processed by the SRAM devices 300, 330 of FIG. 3a or 3b is shown. In the example shown, the SOF field in packet read 704 indicates a new command. Specifically, the first two bits of the CNTL word indicate a READ command, and the remaining 6 bits indicate the SRAM device 300 (FIG. 4) from which the data is to be read. The address fields following the SOF field provide the address in the selected SRAM device 300 to begin reading the data. The SRAM device 300 will initiate a READ at the designated address and continue reading and transmitting the data to the READ output port 314 (FIG. 4) until a STOP READ command is received, or a new READ command is received. The SRAM device 300 resumes the WRITE command where it was interrupted. A SOF field and a CNTL word containing bits set to indicate a STOP WRITE command is included at the end of the packet write 702 to terminate the WRITE command, but the READ command 704 was not terminated and will continue.

[0049] FIG. 8 is an example of a read data stream 800 from two consecutive packet reads that can be output on the read output port 314 of the SRAM devices 300, 330 of FIG. 3a or 3b. The data from each READ command is delimited by a SOF field. The READ command continues to output data until a new read command interrupts the current read command or a STOP READ command is detected by the SRAM device 300, 330. For example, the read data stream 800 shows data from a first READ command, a SOF field to delimit the data from a second READ command, and an IDLE field at the end of the read data stream 800 to indicate that a STOP READ command was detected by the SRAM device 300, 330. Note that the read data stream 800 does not indicate whether a WRITE command was interrupted to perform the READ commands.

[0050] FIG. 9 is an example of a packet write 902 followed immediately by a packet read 904 that can be processed by the SRAM devices 300, 330 of FIG. 3a or 3b. Packet write 902 includes the SOF field to indicate the start of a new control word and initiate the transaction by framing the data and synchronizing the serial bit stream. The CNTL word indicates that the WRITE command is to be performed and the SRAM device 300 in the SRAM series 400, 401 (FIG. 4) into which the data will be written. The ADDR fields indicate the starting address in the specified SRAM device 300 for writing the data. The STOP WRITE command terminates the WRITE command once all of the data has been written to memory.

[0051] The packet read 904 begins with the SOF field to indicate a new command. Specifically, the first two bits of the CNTL word indicate a READ command, and the remaining 6 bits indicate the SRAM device 300 (FIG. 4) from which the data is to be read. The address fields following the SOF field provide the address in the selected SRAM device 300 to begin reading the data. The SRAM device 300 completes the READ command and transmits the data to the READ output port 314 (FIG. 3b). A SOF field is sent to indicate the start of a new transaction, and a STOP READ command in the CNTL word terminates the previous READ command.

[0052] In the example shown in FIG. 9, the minimum READ length is 4 bytes. Five byte transfers are required to initiate a READ command, (i.e., the SOF field, CNTL word, and address bytes in packet read 904). The READ output data will consist of five byte transfers minimum (i.e., the SOF field, and four data bytes), to correspond with the time needed to initiate a new read or a STOP READ command. The WRITE command terminates after the STOP WRITE command is detected, however, the data can be written to memory as the CNTL word in the read packet 904 is parsed. Notably, the data written by packet write 902 can be read by the consecutive packet read 904. Other embodiments may require longer or shorter minimum reads, depending upon the number of bytes of address required to complete a READ command, which can vary based on the amount of memory and the address granularity.

[0053] Referring now to FIGS. 10a and 10b, FIG. 10a shows an example of two minimum-sized sequential read packets 1002, 1004 to be input to the write input port 302 of the SRAM devices 300, 330 of FIG. 3a or 3b. FIG. 10b shows the output on the read output port 314 resulting from performing the read commands in the packet reads 1002, 1004 of FIG. 10a. Each packet read 1002, 1004 includes a minimum of four byte transfers (SOF field, CNTL word, and the address from which data is to be read). During frames 1-8, the first READ command is performed, and the beginning of the second READ command is beginning, as shown in FIG. 10a. The data from the first READ command becomes available after the 9th (ninth) frame, as shown in FIG. 10b. This example of sequential read packets 1002, 1004 shows the worst case time delay for the READ command.

[0054] Referring to FIGS. 3a, 3b,and 5, in some embodiments, to provide traceability of the read data, an extra CNTL word can be added to the data packet 500 to identify the data read from the SRAM module 308. The extra CNTL word can be used to indicate to the SRAM device 300, 330 that a data identifier is included in the packet, for example, following the read address. The identifier can be returned with the read data, and the amount of packet space allocated for the identifier, such as 16 bits, can be selected to meet the requirements of the system. In alternative embodiments, an identifier can be included in the data and the components of the system can be implemented to recognize the identifier by its format and/or location in the data stream.

[0055] Advantageously, some embodiments of SRAM device 300, 330 can write packets as small as one byte by performing a write request (5 bytes) and supplying one byte of data before the next write request. Read requests output a minimum of 5 bytes due to the fact that 5 bytes of address and control information are required to make the request on the write input port 302. Unused data can be discarded.

[0056] Referring to FIGS. 11a and 11b, FIG. 11a shows an example of a known high speed router 1102 that indicates the input and output bandwidths that can be achieved by router 1102. In the example shown, the router 1102 has an internal storage capacity of up to 300 milliseconds, with input and output bandwidths up to 10 gigaHertz for 10 serial input lines at speeds of 1 gigaHertz each.

[0057] FIG. 11b indicates the input and output bandwidths that can be achieved by a high speed router 1104 configured with SRAM devices 300, 330 (FIGS. 3a, 3b) in accordance with the present invention. The router 1104 has an internal storage capacity (in time) similar to router 1102 in FIG. 11a, however, router 1104 achieves input and output bandwidths up to 100 gigaHertz for 10 serial input lines at speeds of 3.125 gigaHertz each. In some embodiments, routers 1102, 1104 can store 300 milliseconds of data before they overflow, however, the internal memory storage capacity of router 1104 (FIG. 11b) is an order of magnitude larger than router 1102 (FIG. 11a).

[0058] Referring again to FIGS. 3a, 3b, and 4, SRAM devices 300, 330 configured in SRAM series 400, 401 can be used to provide cost effective and scalable memory for high-speed routers 100 and other equipment that utilizes SRAM. SRAM devices 300, 330 achieve efficient data transfers with even very small data packets (e.g., 4 byte packet read). Further, the storage capacity of SRAM series 400, 401 can be increased without requiring an increase in the number of pins on ASIC 202 and switch 208.

[0059] The ability to scale the memory capacity and bandwidth while using the same ASIC 202 and switch 208 alleviates problems associated with integrating new ASICS having an increased number of pins for more bandwidth, as found in the prior art. The SRAM series 400, 401 allow additional SRAM devices 300 to be added to SRAM series 400, 401 without incurring board layout problems. Such problems can include attempting to position components to avoid long data bus lines, routing wide data bus lines to memory and processor devices with a large number of pins, and multidrop connections on all busses.

[0060] The requirement of transferring a minimum of five bytes to initiate a READ or a WRITE command allows the internal architecture of SRAM devices 300, 330 to process READ commands and WRITE commands in parallel. As a result, the internal bandwidth of the SRAM devices 300, 330 can be greatly reduced from the external bandwidth. This allows a significant reduction in power dissipation and a reduction in the required internal memory access speed. Embodiments of SRAM devices 300, 330 which require additional latency from the initiation of the READ command to the availability of the data, could be developed using DRAM technology.

[0061] SRAM devices 300, 330 configured in SRAM series 400, 401 also alleviate address fan-out problems because each SRAM series 400, 401 has only one write input port 402 and one read output port 404 that interfaces with ASIC 202. Further, SRAM series 400, 401 alleviate noise issues that arise with heavy bus loading and drivers, line termination, bus contention, and long data bus lines found in the prior art. A further advantage with SRAM series 400, 401 is that there are no logistical problems with bus arbitration, transferring small amounts of data on very wide busses, bottlenecks at the memory interface, and bus/clock skew and bit alignment.

[0062] One skilled in the art will recognize the many teachings of this disclosure and be able to apply these teachings to additional embodiments and, indeed, to other kinds of devices, as well, without departing from the teachings of this disclosure. Consequently, the scope of the invention should not be inferred as being limited by the exemplary embodiments described herein, but rather should be viewed as teaching in the art far greater than just these exemplary embodiments. Accordingly, other embodiments, variations, and improvements not described herein are not necessarily excluded from the scope of the invention.