CS311 Lecture: Memory Devices last revised 10/17/11Need: Projectables: Dynamic RAM cell, ROM Basic Cell, EPROM Basic Cell, Disk headIntroduction------------ A. We already seen how many different types of information can be represented by appropriate binary bit patterns. This allows computer systems to store and process numbers, text, graphics, sounds, and video. B. In fact, the ability to store and transmit various kinds of information has become of increasing importance. Many computers are used primarily for information storage and retrieval, rather than for numeric computation. (Recall the homework problem asking about what name might have been given to "computers" if they were developed today.) This has been made possible, in part, by dramatic increases in the capacity of computer storage media. C. In the VonNeumann computer model, one of the major components was the memory. We now consider various technologies that are used for this purpose in modern computers. D. At the present state of the art, five broad kinds of technology are used for in common use for memory. Each technology can be characterized by several parameters: 1. Cost per bit 2. Access time - the delay between the time a particular item is requested and the time it becomes available. a. For some kinds of memory (random access), the access time is the same for all locations in the memory. b. For others (direct access) any item can be accessed, but there is a time penalty for moving around from one location to another. c. For still others (sequential access) it is most practical to process the data in a certain fixed order. (Direct access is possible but very costly in terms of time.) 3. Cycle time - this includes access time plus any "recovery" time needed before accessing the next item. It turns out that fast access/cycle time and high cost per bit go hand in hand, which means that most practical systems will use several different technologies with relatively small quantities of the faster technologies and larger quantities of the slower ones. 4. Volatility: is the contents lost when the power is turned off (or in case of a power failure)? E. The major technologies are these (in order of decreasing speed and cost per bit). Actually, the last category is used only for long-term storage of files, not as part of the memory system per se.1. The CPU itself contains a small amount of memory in the form of internal registers, realized by flip-flops. Many modern CPUs also have a small amount of "addressable" memory (memory accessed by operations like lw and sw on mips) on the CPU chip, realized the same way. Registers and on-chip memory are, of course, as fast as the CPU itself; but one cannot purchase more of these at any price, since their quantity is fixed by the CPU architecture. a. Typical cost per bit: not applicable (can't buy more) b. Typical access/cycle time: well under 1 ns (must be much less than 1 / CPU clock rate) c. Volatile 2. Separate semiconductor memory chips. These are of several different types, as we shall see shortly, and are used to implement what we commonly call MAIN MEMORY - the memory where a program's code and variables are normally stored - as well as CACHE MEMORY (which we will discuss in conjunction with memory hierarchies). a. Typical cost per bit: $.001 to $.000000001 ($1e-9) b. Typical access/cycle time: i. Access time ranges from a few ns on up, depending on type. ii. For some types, cycle time is about twice access time. c. Some types are volatile; some are read-only and hence nonvolatile. 3. Magnetic disk is relatively inexpensive compared to the other technologies we have listed, but also has an access time that is 4-5 orders of magnitude slower than semiconductor memory. It also has the advantage of being non-volatile; if system power is turned off, its contents are not lost. a. Magnetic disk is typically used for two purposes: i. For long-term storage of files. ii. As an extension of main memory (virtual memory) - which we will discuss in conjunction with memory hierarchies. b. Typical cost per bit: $.0000000001 ($1e-10) c. Typical access/cycle time: 10-20 milliseconds - varies with location on the disk (direct access but not random access) d. Not volatile 4. Optical disk (CD, DVD) and magnetic tape are used for long-term archival storage of data, but not as part of the memory hierarchy per se. a. Typical cost per bit: $.0000000001 ($1e-10) (cost of removable media - does not include cost of drive) b. Typical access time: 100's of ms for optical disks; seconds for tape - most suitable for sequential access c. Not volatile. Some media (eg. regular CD, DVD) are read only. F. We will discuss two of these technologies now: semiconductor memory, and magnetic disks. We discussed CPU registers when we discussed the CPU, and will discuss on the CPU chip cache when we discuss memory hierarchies. Due to time limitations, we will not discuss optical disks or magnetic tape further.I. Overview of Semiconductor Memory Devices- -------- -- ------------- ------ ------- A. Generally, a computer system will have from one to several hundred semiconductor memory chips. (However, there are CPU chips available which have a complete memory system right on the chip, producing a monolithic computer for use in equipment controllers etc.) For now, though, we focus on separate memory chips. 1. A memory chip is characterized by: a. Its CAPABILITIES and TECHNOLOGY i. ROM - read only memory (may be factory programmed, field programmable, field programmable/eraseable, or electrically alterable.) ii. RAM - read/write memory (may be dynamic or static). [ Note: the terminology is a misnomer. ROM is also random access! ] iii. We will cover memory technologies in