system memory之Memory Access and Access Time

Memory Access and Access Time

When memory is read or written, this is called a memory access. A specific procedure is used to control each access to memory, which consists of having the memory controller generate the correct signals to specify which memory location needs to be accessed, and then having the data show up on the data bus to be read by the processor or whatever other device requested it.

In order to understand how memory is accessed, it is first necessary to have a basic understanding of how memory chips are addressed. Let's take as an example a common 16Mbit chip, configured as 4Mx4. This means that there are 4M (4,194,304) addresses with 4 bits each; so there are 4,194,304 different memory locations--sometimes called cells--each of which contains 4 bits of data.. 4,194,304 is equal to 2^22, which means 22 bits are required to uniquely address that number of memory locations. Thus, in theory 22 address lines are required.

However, in practice, memory chips do not have this many address lines. They are instead logically organized as a "square" of rows and columns. The low-order 11 bits are considered the "row" and the high-order 11 bits the "column". First the row address is sent to the chip, and then the column address. For example, let's suppose that we want to access memory location 2,871,405 in this chip. This corresponds to a binary address of "10101111010 00001101101". First, "00001101101" would be sent to select the "row", and then "10101111010" would be sent to select the column. This combination selects the unique location of memory address 2,871,405. This is analogous to how you might select a particular cell on a spreadsheet: go to row #34, say, and then look at column "J" to find cell "J34".

Intuitively, it would seem that designing memory chips in this manner is both more complex and slower than just putting one address pin on the chip for each address line required to uniquely address the chip--why not just put 22 address pins on the chip? It may not surprise you to learn that the answer is "cost". By using the row/column method, it is possible to greatly reduce the number of pins on the DRAM chip. Here, 11 address pins are required instead of 22 (though you lose a small part of the "savings" of 22-11=11 to additional control signals that are needed to manage the row/column timing.) You also save some of the buffers and other circuitry that are required for each address line. Certainly having to send the address in two "chunks" slows down the addressing process, but keeping the chip smaller and with fewer inputs allows it to use less power, which makes it possible to run the chip faster, partially offsetting the loss in access speed.

Of course, a PC doesn't have a single memory chip; most have dozens, depending on total memory capacity and the size of DRAMs being used. The chips are arranged into modules, and then into banks, and the memory controller manages which sets of chips are read from or written to. Since a modern PC reads or writes 64 bits at a time, each read or write involves simultaneous accesses to as many as 64 different DRAM chips.

Here is a simplified walkthrough of how a basic read memory access is performed. This is a conventional asynchronous read, because the timing signals are not tied to the main system clock; synchronous DRAM uses different timing signals:

1. The address for the memory location to be read is placed on the address bus.
2. The memory controller decodes the memory address and determines which chips are to be accessed.
3. The lower half of the address ("row") is sent to the chips to be read.
4. After allowing sufficient time for the row address signals to stabilize, the memory controller sets the row address strobe (sometimes called row address select) signal to zero. (This line is abbreviated as "RAS" with a horizontal line over it. The horizontal line is a short-hand code that tell engineers working with the circuit that the signal is "active low", meaning that the chip is looking for it to be set to zero as a signal to "do something". There's no way in HTML to reliably use this notation so instead, I will write "/RAS".)
5. When the /RAS signal has settled at zero, the entire row selected (all 2^11 columns in the example above, or 2048 different cells of 4 bits each) is read by the circuits in the chip. Note that this action refreshes all the cells in that row; refreshing is done one row at a time.
6. The higher half of the address ("column") is sent to the chips to be read.
7. After allowing sufficient time for the column address signals to stabilize, the memory controller sets the column address strobe (or column address select) signal to zero. This line is abbreviated as "CAS" with a horizontal line over it, or "/CAS".
8. When the /CAS signal has settled at zero, the selected column is fed to the output buffers of the chip.
9. The output buffers of all the accessed memory chips feed the data out onto the data bus, where the processor or other device that requested the data can read it.

Note that this is a very simplified example, since it doesn't mention all of the various timing signals, and it also ignores common performance enhancements such as multiple-banked modules, burst mode, etc. A write process is performed similarly, except of course that the data is read into the chips instead of being sent out by them. A special signal called "R/W" (actually written with a horizontal line over the "W") controls whether a read or write is being performed during the access.

The amount of time that it takes for the memory to produce the data required, from the start of the access until when the valid data is available for use, is called the memory's access time, sometimes abbreviated t_AC. It is normally measured in nanoseconds (ns). Today's memory normally has access time ranging from 5 to 70 nanoseconds. This is the speed of the DRAM memory itself, which is not necessarily the same as the true speed of the overall memory system. Note that much of the difference in access times of various DRAM technologies has to do with how the memory chips are arranged and controlled, not anything different in the core DRAM chips themselves.

Next: Asynchronous and Synchronous DRAM