Computer memory systems

1
Computer memory systems

• Computer Architecture
• Lecture Notes
• May 30, 2006

2
Memory hierarchy

• Collection of many different devices with
  different physical characteristics and modes of
  operation
• No memory device possesses all the
  characteristics we consider ideal
• Each memory devices has an advantage over the
  others
• Try to gain the advantages of each by combining
  them in one system

3
Characteristics of an ideal memory

• Low cost
• High speed
• High density
• Nonvolatile
• Read/write capable
• Low power
• Durable
• Removable

4
Characteristics of real memory devices

• Several types of memory used in modern computer
  systems
• Most popular types of memory are semiconductor
  chips (ICs) and magnetic and optical media
• Designers need to be familiar with memory types

5
Semiconductor memories

• Possess advantage of speed
• Used for main memory in modern computers
• Magnetic and optical memory used for secondary
  memory
• Dynamic Random Access Memory (DRAM) highest
  information density

6
DRAM

• Consists of large array of capacitors
• Charged capacitor interpreted as storing a binary
  1, uncharged binary 0
• Capacitors drain over time - information must be
  periodically read and then rewritten to memory to
  keep it stored
• Process called dynamic RAM refresh
• Adds to complexity of memory control circuitry

7
DRAM

• Because of memory requirements and amount of DRAM
  that can be fabricated on a single IC, DRAM not
  usually sold in single chip
• Several ICs packaged on a small printed circuit
  board
• Modules come in several forms SIMMs (single
  inline memory modules), DIMMs (dual inline memory
  modules), RIMMs (Rambus integrated memory
  modules)

8
SRAM

• Static random access memory
• Highest-speed semiconductor read / write memory
• Binary information stored as states of latches or
  flip-flops rather than capacitors
• Built in very similar way to registers in the CPU
• Less dense than DRAM
• More expensive to construct
• Volatile like DRAM
• Doesnt require period refreshes like DRAM
• Requires more power for read / write operations
• CMOS (Complementary Metal Oxide Semiconductor)
  requires very little current in standby mode, can
  maintain information for years under battery power

9
Other Memory Devices

• ROM (read-only memory) semiconductor memory
  includes PROMs (programmable read-only memory),
  EPROMs (erasable/programmable read-only memories)
• Roughly comparable to SRAM in cost and density
• Generally operate at slower speeds
• Nonvolatile
• Limitation not writeable
• Used for single-purpose embedded systems, video
  cartidges, BIOS (basic input/output system
  containing bootstrap code))
• A USB flash drive is essentially NAND-type flash
  memory integrated with a USB 1.1 or 2.0
  interface. It is a small, lightweight, removable
  and rewritable data storage device of up to 64
  GB, with the price per MB decreasing rapidly at
  newer, larger capacities. (More information )
  http//simple.wikipedia.org/wiki/USB_Flash_drive

10
Magnetic Memory

• Used for secondary storage (floppy and hard
  disks)
• Access time is longer than semi-conductor memory
• Order of milliseconds or longer
• Tape drives sequential storage device (slow but
  portable), nonvolatile

11
Magnetic Memory (MRAM)

• Operates on principle of magnetoresistance
  electric current used to change magnetic
  properties of a solid-state material
• Pieces of material sandwiched between two
  perpendicular layers of wires, bit is stored at
  each point where one wire crosses another
• To write a bit, current passed through wires,
  changing polarity of the magnet changes the
  electrical resistance of the sandwiched materila
• Reading bit accomplished by passing current
  through the wires connected to the sandwich and
  detecting its resistance high binary one, low
  binary 0
• MRAM nonvolatile

12
MRAM

• Can achieve density, speed, and cost comparable
  to DRAM
• Will replace DRAM
• Instant on computer is possible

13
Optical Memory

• Becoming more and more popular
• Compact disk read-only memory (CD-ROM) common on
  all computers now
• Different types CR-R, CD-RW, DVD-ROMS (digital
  versatile disks)
• Offer most of same advantages (portability,
  nonvolatility, low cost, high density)
• Too slow for main memory writing takes longer
  than reading
• More information about CDs - http//www.usbyte.com
  /common/compact_disk.htm

14
Hierarchical memory systems

• None of the memory systems discussed in the last
  section are ideal
• Each type has certain advantages and
  disadvantages
• Makes sense to use of mix of the different memory
  types

15
Memory hierarchy (modern computer systems)
Most expensive
CPU registers
SmallestCapacity
Level 1 Cache
Fastest
Level 2 cache
Main (primary) memory
Least expensive
Disk (secondary) memory
Highest capacity
Backup storage
Slowest
16
Main Memory Interleaving

• Number of devices needed in order to achieve
  total memory size
• Disadvantages Packaging parts count
• Advantages Fault tolerance, flexibility of
  organization
• Using multiple smaller devices/sets of devices
  allows designer to choose how addressed locations
  are distributed among devices
• Interleaving distribution of memory addresses
  over a number of physically separate storage
  locations
• Type of Interleaving can affect performance of
  memory system

17
High-order interleaving

• Simplest and most common way to organize a
  computers main memory when constructing it from
  a number of smaller devices
• Example - we would feed two lines into a 2-to-4
  decoder, the outputs of which would be connected
  to the Chip Select pins of 4 memory modules.
• Note that this means consecutive addresses are
  stored within the same module, except at the
  boundary. The above arrangement is called
  high-order interleaving, because it uses the
  high-order, i.e. most significant, bits of the
  address to determine which module the word is
  stored in.

18
Low Order Interleaving

• Used to improve bandwidth to a single processor
• Large memory constructed of n number of smaller
  devices
• Difference from H-O interleaving how we map
  memory addresses across different devices or
  groups of devices
• Instead of connecting low-order address bits from
  CPU to all devices in common, we connect
  high-order bits and generate four-chip inputs by
  decoding two lowest order address bits

19
Interleaving

• a) Interleaved Memory A memory is interleaved if
  it is composed of independent modules. If the
  modules can be operated independently and if the
  memory bus line is time shared among memory
  modules, then one should expect an increase in
  the data transfer between the main memory and
  CPU.
• b) High Order Interleaving Consecutive addresses
  in the same memory module.
• c) Low Order Interleaving Consecutive addresses
  in consecutive memory modules.
• Advantages of High Order Interleaving / Low
  Order Interleaving
• Low-order interleaved memory is faster than
  high-order interleaved memory.
• High-order interleaved memory is more suitable in
  transferring a block of information from the
  secondary storage to the main memory.
• High-order interleaved memory is more
  fault-tolerant.
• In a multiprocessor environment, the high -order
  interleaved memory is more efficient.
• Enforcing security and protection is easier in
  high-order interleaved memory, since memory
  blocks can be distributed among processors.

20
L-O Interleaving

• Most important difference between H-O and L-O
  permutation of memory addresses over the several
  devices. Same number of memory locations equally
  divided among devices, addresses are assigned in
  rotation, such that device 0 contains memory
  locations 0,4,8,12. (all those whose binary
  addresses end in 0
• Access sequentially number memory locations, the
  accesses will be distributed over all four
  devices on a rotating basis

21
L-O Interleaving

• Big advantage given some extra hardware (to
  allow separate latching of the addresses and
  transfer of data for each device or banks of
  devices) it is possible to have several memory
  accesses in progress simultaneously
• Allows for significant performance improvement
  over high-order interleaving

22
L-O Interleaving - Disadvantages

• Increase in complexity and cost
• H-O Interleaving is simple since only memory
  access is in progress at a time
• L-O requires either n separate data and address
  buses or to multiplex the addresses and data
  values for up to n simultaneous transactions
  across the same bus
• The additional equipment must be fast (decodes,
  latches, transceivers, etc) have to do n time
  work in the same time may cut into the speed
  gains from L-O interleaving

23
L-O Continued

• L-O interleaving on multi-processor systems
• Memory system designed to maximize bandwidth of
  transfers to or from a single device If one
  processor is taking advantage of accessing
  sequentially numbered memory locations, it is
  using up the full bandwidth, other processors
  required to wait until memory is freed
• Potential remedy Large main memory, use both
  high- and low-order interleaving in same system
• Memory addresses divided into 3 logical parts
  both upper and lower bits externally decoded.
  Upper bits would select an address range composed
  of sets of devices, low-order bits would choose a
  device or set of devices, permuted by address,
  within this larger set middle bits would be
  decoded internally by the devices to select a
  particular location. This combined interleaving
  scheme is the most complex and costly to
  implement. Could be justified in large systems
  where performance is at a premium

24
Logical organization of computer memory

• Bulk of main memory in most computer systems is
  semiconductor RAM (random access memory)
• Not all memory is RAM
• Two other types sequential memories and
  associative memories

25
Random access memories

• Computer programs do not access memory randomly
• What random access means is that any memory
  address can access in the same amount of time,
  regardless of its value
• DRAMs, SRAMs, PROM, EPROM and EEPROM are random
  access memories
• Addressed memory - synonym for RAM
• All RAMs, except ones with addresses for
  individual bits, are accessed by what is called
  word slice. One presents the address i of a word
  and then can read or write all of the bits of
  word i simultaneously
• No mechanism for reading / writing bits from
  different words in one operation (couple be
  helpful for graphics)

26
Orthogonal memory

• Could be used if you wanted to accessed memory by
  either bit-slice or word-slice
• Orthogonal means perpendicular

27
Sequential access memories

• Example tape drive
• Sequential access memories are addressed, in a
  sense, not by absolute address but by relative
  address tells you which way to go to find the
  record

28
Associative memories

• Radically different from RAM and sequential
  access memory
• Also call, content addressable memory (CAM)
• Associative memories identify stored information
  by the actual content that is stored (or at least
  some subset of it)

29
Associative Memory

• Real power and utility of an associative memory
  is the ability to match on a selected part of the
  contents, which is known, in order to obtain the
  related information that is sought
• An associative memory is faster than a software
  search
• 3 required for associative search
• Need to provide an argument or search term
• Need a mask or key that identifies which bit
  positions of the argument to check for a match on
  and which to ignore
• Need some sort of control over conflict
  resolution, or at least a way to detect conflicts

30
Associative Memory

• Associative searches can return a unique match,
  multiple matches, or no match
• Associate memory array
• Argument register A that holds the item to be
  searched for
• Key register K in which bits equal to 1 indicate
  position to check for a match and zeroes denote
  positions to ignore
• Match register M contains one bit for each work
  in the associative array
• If the logical OR of all the match register bits
  is zero, no match found, if 1, at least one match
  was found
• Examining the individual bits of M allows one to
  determine how many matches occurred and in what
  locations

31
Associative Memory

• Construction of registers A, K, and M
  straightforward
• Construction of associative array memory cells
  could be constructed fo capacitors (like DRAM),
  flip-flogs (like SRAM, or some other technology
• Purpose of associative memory be able to
  perform high-speed search of stored information
  assume each bit of data stored in a D flip-flop
  or similar device
• Mechanism for reading/writing these bits is the
  same as it would be with any static Ram to store
  we place it on the D input and clock the
  deviceto read a stored bit we simply look at the
  state of Q output
• Additional logic required to perform search and
  check for matches

32
Associative Memory

• Logic decreases the density of the memory cells
  and add considerably to the cost per bit of
  fabricating the memory but may be worth it in
  terms of speeding up the search for information