The memory controller is a digital circuit which manages the flow of data going to and from the main memory. It can be a separate chip or integrated into another chip, such as on the die of a microprocessor. This is also called a Memory Chip Controller (MCC).[1]
Computers using Intel microprocessors have traditionally had a memory controller implemented on their motherboard's northbridge, but many modern microprocessors, such as DEC/Compaq's Alpha 21364, AMD's Athlon 64 and Opteron processors, IBM's POWER5, Sun Microsystems's UltraSPARC T1, and more recently Intel's Core i7 have an integrated memory controller (IMC) on the microprocessor in order to reduce memory latency. While this has the potential to increase the system's performance, it locks the microprocessor to a specific type (or types) of memory, forcing a redesign in order to support newer memory technologies. When DDR2 SDRAM was introduced, AMD released new Athlon 64 CPUs. These new models, with a DDR2 controller, use a different physical socket (known as Socket AM2), so that they will only fit in motherboards designed for the new type of RAM. When the memory controller is not on-die, the same CPU may be installed on a new motherboard, with an updated northbridge.
The integration of the memory controller onto the die of the microprocessor is not a new concept. Some microprocessors in the 1990s such as the DEC Alpha 21066 and HP PA-7300LC had integrated memory controllers, but rather than for performance gains, this was implemented to reduce the cost of systems by eliminating the need for an external memory controller.
Contents |
Memory controllers contain the logic necessary to read and write to DRAM, and to "refresh" the DRAM by sending current through the entire device. Without constant refreshes, DRAM will lose the data written to it as the capacitors leak their charge within a fraction of a second (not less than 64 milliseconds according to JEDEC standards).
Reading and writing to DRAM is performed by selecting the row and column data addresses of the DRAM as the inputs to the multiplexer circuit, where the demultiplexer on the DRAM uses the converted inputs to select the correct memory location and return the data, which is then passed back through a multiplexer to consolidate the data in order to reduce the required bus width for the operation.
Bus width is the number of parallel lines available to communicate with the memory cell. Memory controllers' bus widths range from 8-bit in earlier systems, to 512-bit in more complicated systems and video cards (typically implemented as four 64-bit simultaneous memory controllers operating in parallel, though some are designed to operate in "gang mode" where two 64-bit memory controllers can be used to access a 128-bit memory device).
Double Data Rate DDR memory controllers are used to drive DDR SDRAM, where data is transferred on the rising and falling access of the memory clock of the system. DDR memory controllers are significantly more complicated than Single Data Rate controllers, but allow for twice the data to be transferred without increasing the clock rate or increasing the bus width to the memory cell.
Dual Channel memory controllers are memory controllers where the DRAM devices are separated on to two different buses to allow two memory controllers to access them in parallel. This doubles the theoretical amount of bandwidth of the bus. In theory, more channels can be built (a channel for every DRAM cell would be the ideal solution), but due to wire count, line capacitance, and the need for parallel access lines to have identical lengths, more channels are very difficult to add.
Fully buffered memory systems place a memory buffer device on every memory module (called an FB-DIMM when Fully Buffered RAM is used), which unlike traditional memory controller devices, use a serial data link to the memory controller instead of the parallel link used in previous RAM designs. This decreases the number of the wires necessary to place the memory devices on a motherboard (allowing for a smaller number of layers to be used, meaning more memory devices can be placed on a single board), at the expense of increasing latency (the time necessary to access a memory location). This increase is due to the time required to convert the parallel information read from the DRAM cell to the serial format used by the FB-DIMM controller, and back to a parallel form in the memory controller on the motherboard. In theory, the FB-DIMM's memory buffer device could be built to access any DRAM cells, allowing for memory cell agnostic memory controller design, but this has not been demonstrated, as the technology is in its infancy.