Serial and Parallel Communication

With eight wires, using binary on each line, you can send one BYTE at a time.

A Relay Station

Wire degrades the signal. Need a human or electro-mechanical relay every 50 or so miles.

An electro-mechanical relay is a switch operated by an electro-magnet.

It keeps our binary digits clean.

Buffering (relaying) the telegraph signal:

Viewing it as a logic circuit.

Rather than many miles, the same ideas are used for sending information a few centimetres over a circuit board or a fraction of a millimetre over a silicon chip. We then need only one battery or power supply.

Mechanical Teletype: Advantages: don't need to learn Morse Code, don't need to manually write it down at the receiving station, faster.

Modern character set:

Use an 8-bit byte for each character sent.

Compatibility: You can still plug a 1950 mechanical teletype into a recent laptop using the RS-232 serial port.

Only one wire in each direction is actually needed: so with a ground (earth reference) connection as well, a total of three pins are commonly used on the nine-pin serial connector.

Gate Transfer Curves: Keep those digits clean!

Plots of input and output voltage for the basic components.

Characteristic shape: flat at the edges, sharply changing in the middle: keeps binary digits clean.

Voltage meanings and margins (five volt system):

How To Store One Bit Electrically

Electro-mechanical implementation: Using a relay

Alternative Implementation: Electronic

Flip-flops made from two inverters in a ring are the main means of making registers in silicon chips today.

The mechanical and electronic bistables are both volatile, meaning that information is lost when the power is removed.

Speed of switching: relay is at least a million times slower than a transistor.

Size of the switch: relay uses about 100 million times the substrate area.

Besides bistables, most computers today store data as magnetic spots on the hard disk and static electricity charges in Flash and DRAM memories. These give a few orders of magnitude greater density than transistor bi-stables.

A Logic Gate

The simplest function we can compute is implemented by a basic logic gate.

My logic gate is implemented with two relays: while they are both activated by logic ones on their inputs, the output becomes a logic one. On silicon chips, gates are made out of field-effect transistors and four transistors are used instead of two relays.

Other Basic Gates

There is also an XOR gate, and versions of AND and OR with inverted outputs called NAND and NOR.

Adder Circuits

A half adder tallies two inputs creating a binary number in the range 0 to 2.

A B | C S -------------- 0 0 | 0 0 0 1 | 0 1 1 0 | 0 1 1 1 | 1 0

The full adder: made of two half adders:

A full-adder tallies three inputs creating a binary number in the range 0 to 3.

A full-adder tattoo:

Larger adders can be made by cascading full adders:

A three bit-adder is made from three full adders (six half adders).

A3 A2 A1 B3 B2 B1 | C4 S3 S2 S1 ------------------------------------- 1 0 1 0 1 0 | 0 1 1 1 5 + 2 = 7

Three-bit adder schematic symbol (hides the details):

Gate count: each gate is roughly 4 transistors. A three bit adders is 15 gates = 60 transistors.

Moore's Law

Moore's law describes an important trend in the history of computer hardware: that the number of transistors that can be inexpensively placed on an integrated circuit is increasing exponentially, doubling approximately every two years. The observation was first made by Intel co-founder Gordon E. Moore in a 1965 paper.

The largest chips are about the size of a fingernail.

There are five or six major chips in a Nintendo DS, and dozens of small ones.

Internal parts of a smart phone (from Modern SoC Design on ARM by DJ Greaves (P) 2021); full text (local access only) LINK).

Multiplexor: Another Basic Component

Making a transparent latch flip-flop

Putting flip-flops together to make a register.

(These flip-flops are edge-triggered, meaning that they sample their input data on the positive edge of the clock. Internally, they use a pair of transparent latches).

Primary Storage/Main Memory: RAM - Random Access Memory

Possible implementation of RAM using a linear array of registers:

Dynamic RAM

Larger arrays are made using dynamic technology, using electric charge on a capacitor to store a bit instead of the ring-pair of inverters: .

Other important components: Counter and Clock

The clock generates a regular series of transitions.

A counter increments on each clock pulse.

How to make a counter from an adder and a broadside register:

With ten output wires, the count range is 0 to 4095.

Schematic symbol hides the details: .

How Computers Work - Lecture Two - We Now Have Sufficient Components to Build a Complete Computer

Come back and look at these slides in your own time ... or perhaps study the Little Man Computer

or perhaps Nand To Tetris: The Elements of Computing Systems / Nisan & Schocken www.nand2tetris.org"

Counter Connected to a RAM memory

The counter output can index a RAM containing instructions.

Put instructions in the RAM and it steps through them in sequence, like stepping through the Jacquard punch cards.

A jumpable counter

Here we add a multiplexor between the adder and the broadside register.

This enables a new value to be seeded.

One of the instructions can be a GOTO, that operates the multiplexor to select a branch destination.

An Execution Unit with Control Unit

The ALU is an arithmetic and logic unit: it can compute a variety of functions of its two inputs.

Combining the Execution and Control Units

The combination of control and execution units on a single chip is a called a microprocessor.

[When there are several of these on a chip, each pair of them is called a core and the core also typically has its own small cache.]

A single bus connects the processor to the memory system. This has been called the von-Neumann bottleneck. It is avoided in today's computers by dedicating a large fraction of the processor chip to cache memories. Having simultaneous access to several caches gives access to instructions and data in parallel.

(You can do all this using relays if you want: A relay-based computer).

Our Program Again

Here is our program again, together with the recap of the blocks found in an everyday (von Neumann) computer:

Computer Architecture (further reading):

Understanding typical computer bus structures:

The Future

Conventional computers cannot be made any smaller or faster out of silicon: we are running out of atoms!

Today's (2014) new computers have multiple conventional computer cores on one chip: for example the "Intel Core 2 Duo" or the AMD "Turion". The cores all sharing the same main memory.

Since data movement has become the dominant use of electricity in computers, and the cost of ALU logic has become close to zero with increasingly smaller transistors, there is also movement back towards the style of Babbage's Difference Engine. In that design, functionality was hardwired and ALUs were associated with storage locations instead of being centralised in the mill. Custom processors for social graph processing, AI inference and DNA sequencing are very common (2022).

But the mainstream von Neumann design remains king and we are seeing the number of cores on the processor chip growing. New classes of algorithms that are better adapted for parallel processing are needed (2022).

But the size of an atom and the speed of light are fixed, so a new approach is needed for further performance improvements ... ?

Thank you for listening.

Computer Hardware: How Computers Work

Dr David Greaves

Corpus Christi College

and

Computer Laboratory

University of Cambridge