WHY IS A BYTE 8 BITS? OR IS IT?

I recently received an e-mail from one Zeno Luiz Iensen Nadal, a worker for Siemens in Brazil. He asked "My Algorythms teacher asked me and my colleagues 'Why a byte has eight bits?' Is there a technical answer for that?"

Of course I could not resist a reply to someone named Zeno, after that teacher of ancient times. Some people copied on the reply thought it a useful document, so (having done the hard work already) I add it to my site as further bite of history.

The answer is that some do, and some don't. But that takes explaining, as follows:

If computers worked entirely in binary (and some did a long time ago), and did nothing but calculations with binary numbers, there would be no bytes.

But to use and manipulate character information we must have encodings for those symbols. And much of this was already known from punch card days.

The punch card of IBM (others existed) had 12 rows and 80 columns. Each column was assigned to a symbol, a term I use here although they have fancier names nowadays because computers have been used in so many new ways.

The columns, going down, starting from the top, were 12-11-0-1-2-3-4-5-6-7-8-9. A punch in the 0 to 9 rows signified the digits 0-9. A group of columns could be called a "field", and a number in such a field could carry a plus sign for the number (an additional punch in top row 12 of the units position of the number), or a minus sign (an additional punch in row 11 just under that).

Then they started to need alphabets. This was accomplished by adding the 12 punch to the digits 1-9 to make letters A through I, the 11 punch to make letters J through R. For S through Z they added the 0 punch to the digits 2 through 9 (the 0-1 combination was skipped -- 3x9=27, but the English alphabet has only 26 letters). The 12, 11, and 0 punches were called "zones", and you'll notice them today lurking in the high-order 4 bits. Remember that this was much prior to binary representations of those same characters.

The first bonus was that the 12 and 11 punches without any 0-9 punch gave us the characters + and -. But no other punctuation was represented then, not even a period (dot, full stop) in IBM or telecommunication equipment. One can see this in early telegrams, where one said "I MISS YOU STOP COME HOME STOP". "STOP" stood for the period the machine did not have.

Then punctuation and other marks had combinations of punches assigned, but there had to be 3 punches in a column to do this. In most case the third punch was an extra "8".

In this way, with 10 digits, 26 alphabetic, and 11 others, IBM got to 47 characters. UNIVAC, with different punch cards (round holes, not rectangles, and 90 columns, not 80) got to about 54. But most of these were commercial characters. When FORTRAN came along, they needed, for example, a "divide" symbol, and an "=" symbol, and others not in the commercial set. So they had to use an alternate set of rules for scientific and mathematical work. A set of FORTRAN cards would cause havoc in payroll!

With many early computers these punch cards were used as input and output, and inasmuch as the total number of characters representable did not exceed 64, why not use just 6 bits each to represent them? The same applied to 6-track punched tape for teletypes.

In this period I came to work for IBM, and saw all the confusion caused by the 64-character limitation. Especially when we started to think about word processing, which would require both upper and lower case. Add 26 lower case letters to 47 existing, and one got 73 -- 9 more than 6 bits could represent.

I even made a proposal (in view of STRETCH, the very first computer I know of with an 8-bit byte) that would extend the number of punch card character codes to 256 [1]. Some folks took it seriously. I thought of it as a spoof.

So some folks started thinking about 7-bit characters, but this was ridiculous. With IBM's STRETCH computer as background, handling 64-character words divisible into groups of 8 (I designed the character set for it, under the guidance of Dr. Werner Buchholz, the man who DID coin the term "byte" for an 8-bit grouping). [2] It seemed reasonable to make a universal 8-bit character set, handling up to 256. In those days my mantra was "powers of 2 are magic". And so the group I headed developed and justified such a proposal [3].

That was a little too much progress when presented to the standards group that was to formalize ASCII, so they stopped short for the moment with a 7-bit set, or else an 8-bit set with the upper half left for future work.

The IBM 360 used 8-bit characters, although not ASCII directly. Thus Buchholz's "byte" caught on everywhere. I myself did not like the name for many reasons. The design had 8 bits moving around in parallel. But then came a new IBM part, with 9 bits for self-checking, both inside the CPU and in the tape drives. I exposed this 9-bit byte to the press in 1973. But long before that, when I headed software operations for Cie. Bull in France in 1965-66, I insisted that "byte" be deprecated in favor of "octet".

You can notice that my preference then is now the preferred term. It is justified by new communications methods that can carry 16, 32, 64, and even 128 bits in parallel. But some foolish people now refer to a "16-bit byte" because of this parallel transfer, which is visible in the UNICODE set. I'm not sure, but maybe this should be called a "hextet".

But you will notice that I am still correct. Powers of 2 are still magic!

REFERENCES

1. R.W.Bemer, "A proposal for a generalized card code of 256 characters",
   Commun. ACM 2, No. 9, 19-23, 1959 Sep
   -- Computing Reviews 00025
   Early public hint of 8-bit bytes to come.
   
2. R.W.Bemer, W.Buchholz, "An extended character set standard",
   IBM Tech. Pub. TR00.18000.705, 1960 Jan, rev. TR00.721, 1960 Jun
   -- Computing Reviews 00813
   
3. R.W.Bemer, H.J.Smith, Jr., F.A.Williams,
   "Design of an improved transmission/data processing code",
   Commun. ACM 4, No. 5, 212-217, 225, 1961 May
   -- Computer Abstracts 61-1920
   ASCII in its original form.