US Flag - fm

Information Technology (IT) Pioneers

Retirees and former employees of Unisys, Lockheed Martin, and their heritage companies

 36bit Computers, Chapter 56

1. Introduction

The first 36-bit machine was the ERA 1103 (Atlas II). The 1103A was a technology upgrade to be called the UNIVAC SCIENTIFIC. This newsletter dated February 28, 1953 has an artist's drawing of the 1103.

The 1104 used 1103A technology with a 30-bit ISA thus isn't included on this page.  The 1105 was a technology upgrade of the 1103A, leading to the 1107 series as described in the Processors, Commercial chapter. A video summary runs about 15 mins, presented by the Computer History Archives Project (CHAP),

2. 36-bit Computer Descriptions

Many thanks to the Bit Savers website, they have a lot of document information about these early systems. Read the History and Evolution of the 1100-2200 mainframe computers written by Unisys Fellow Richard Petschauer.

2.1 1103 and 1103A 

By Harry Wise - The 1103 stepped up to a 36 bit word length. The difference - the 24 bit ISA contained an instruction and the drum address of the operand. The 36 bit ISA had an instruction, the operand address, and the address of the next instruction. This innovation meant that the assembler could strategically place the next instruction around the drum so that it would be under the read heads at the instant that the current instruction finished execution. These machines used logic implemented with vacuum tubes. After delivering twenty-seven 1103 based systems, one 1104 system, and ten 1105 drum based systems, UNIVAC began the 1100 commercial series with the 1107 transistor/core memory based system and a different ISA that had some characteristics of the 30-bit NTDS computer architecture.

George Gray and Ron Q. Smith of Unisys co-authored three articles on the history of Sperry computers published in the IEEE Annals of the History of Computing. George provided the following 1103 and 1105 customer information. [lab]

S/N Customer S/N Customer
1 Consolidated Vultee Aircraft  2 U.S. Air Force-Eglin AFB 
3 White Sands Missile Range 4 Ramo Wooldridge, then U of Minnesota in March of 1958
5 Westinghouse Electric 6 Johns Hopkins University - ORO
7 U.S. Air Force-Wright Center 8 NACA Lewis Research Center 
9 UNIVAC St. Paul Engineering     
10  Lockheed Aircraft  11 Boeing Aircraft
12 U.S. Air Force Missile Center  13 Johns Hopkins University - APL
14 U.S. Army Engineer Laboratory then Johns Hopkins Univ. - ORO    
15 Ramo Wooldridge 16 Westinghouse Electric
17 Johns Hopkins University - ORO 18 U.S. Army Signal Corps
19 White Sands Missile Range 20  U.S. Air Force Missile Center
22 Lockheed Missile Division 23 U.S. Air Force – Wright Center
24 U.S. Navy  26  U.S. Navy 
27 Lockheed Missile Division    S/N 21 and S/N 25 aren't known by our information sources


2.2 1105

This 1955 System Console photo is from the 'Bit Savers' collection. LABenson recognizes the person at the console as Bill Klingner who went on to form his own company, Northport Engineering. I met Bill in the late 80s when he and I were in the same investment club.

S/N Customer   S/N Customer
1,2 U.S. Bureau of the Census     3   U. of North Carolina  
4 U.S. Air Force - Air Materiel      5  Illinois Institute of Technology 
6,9 U.S. Air Force - ROAMA    7,8  U.S. Air Force – Supply Center 
10 Government of Brazil       

 Thanks to George Gray for this customer delivery information.

2.3 CP-667 - by Jim Rapinac, Lowell Benson, Larry Bolton, Jorgen Andersen, and Larye Parkins.

The CP-667 was a new design based on Don Ream's NTDS requirements design, it wasn’t based on a commercial Univac computer. It was slated for use on the Navy's Mobile OPCON program which was subsequently canceled. Not sure what happened to this program. Only Eric Swenson, Vern Leas, and Don Ream knew.

       The cabinet was based on the AN/USQ-20 cabinet because I moved many basic Q20 mechanical components from the Q20 crib to the CP-667 crib! Some components like gussets and side plates were new designs because the CP-667 cabinet was slightly larger than the Q20 due to memory size, 16 I/O channels, thin film memory, etc. Mechanical Design Engineers included Les Nesler, Seri Yermolenko, and Tony Just, who was also involved in the circuit design. Hy Osofsky and Glen Kregess were the key logic designers, shown here with a CP-667, circa 1963. They reported to Manager Finley McLeod who worked for Director George Raymond.

I was the PERT planning engineer for the CP667 and we only built one computer, much to our disappointment. I placed the first Purchase Order (PO) for a wire wrap machine for the back panels. The CP-667 was years ahead in technology. This design was the first UNIVAC computer to use Integrated Circuits (ICs) which were packaged in TO-5 cans – looked like big transistors but had 10 leads instead of three.
  The main memory used a new core, smaller than previous cores, which we obtained from General Ceramics, Keasby, NJ. I negotiated the price and delivery terms for this smaller core and the memory planes used a new core stringing machine.
  Also, the CP-667 was the first product to use the automatic wire wrap machine. Larry Reid was the Mfg. Director and he was reluctant to sign the PO but finally he did.  Regards; Rapp
Photo at the right came from Curt Nelson’s files. [lab]

Finley McLeod led the design team developing the CP-667. It had 131k of directly addressable core memory. Although it was designed as a 36-bit machine, the flip of a switch on the control panel had this unit revert to 30-bit operation.

I, Lowell, was a computer operator in plant 1 from 1963 to '66. We had a CP-667 there and I ran a few programs on it. My memory is that I was told that there was also one in a center in San Diego plus one that was supposed to go aboard a ship - but damaged in a fire thus became spare parts. Because of my recollection and that the Sperry Univac Computer Genealogy chart showing the number (3), I was thinking that Boslaugh's book with qty 1 CP-667 is in error. There was just one for the Navy when the 2nd was in the fire.  The third stayed in St. Paul, used for a variety of internal projects.

This machine and the one from San Diego were in the Eagan METC during the 80s according to Gene Achterberg and Dr. Carl Glewwe. Carl also told me that since they couldn’t sell these computers, his job was to use them on various projects to recover some of the costs. David P. Andersen told me that some of their early Voice research software used the CP-667 before they transitioned to the 1616 mini-computers. Earl Vraa chipped in that the early Computer Aided Design work used the CP-667 computer. [LABenson]

According to one of my listings, the CP-667 used DTL gates manufactured using hybrid techniques. Those were contained in 10-lead TO-5 cans. The CP-667 Diode Hybrid Gates were documented in procurement specs numbering 7900032 through 7900043. [Larry Bolton]

There are six of these TO-5 cans on the 27-pin card shown on the right. [lab]

[from Larry Bolton] I found the exact card you have the photos of in our archives. It is a 4224080 card and has six gold colored 10-lead TO-5 can devices on it. I looked at the parts and got part numbers. There are three Motorola SC90 parts, one Motorola SC92 part, and two Motorola SC94 parts. All were made in 1963.
I went to the vault and got the microfilm for what the cross reference chart mentioned above said were the procurement specs for the CP-667 hybrids. I found three drawings. The 7900034 shows that a Motorola SC92 is the source, the 7900035 shows that a Motorola SC93 was the source, and the 7900036 shows that a Motorola SC94 was the source. By extension, it can be assumed that the SC90 would have been the source for a 7900032. These drawings were generated to support Lee Granberg in 1962. The SC92 device is nine 1N914 discrete devices along with four 1K resistors in the form of a 1-1-1-2 diode gate. The SC94 device is two separate 2N2501 transistors, each having an input diode and two bias resistors. This is called a two transistor gate. I could not determine the configuration for the SC90 since the corresponding 7900032 drawing had been superseded and apparently purged from the system. The superseding part probably has more information but the microfilm reel is in a format that the vault reader could not load. According to the chart, the SC90 would be similar to the SC92 device but in a 3-4 diode gate configuration. So, without a question, this particular card was made using hybrid devices. {Editor's note: It is quite likely that in 1963 the vernacular was that these multiple devices configured into the TO-05 cans were called integrated circuits. They were not the 'monolithic integrated circuits' which were used in the 1824 design in 1965 nor in the CP-901 design in 1967.} In order to verify this, I went to the computer system to get a parts list for the 4224080 card. To my surprise, none of these parts were on the parts list. Instead, the parts list has three 7901003 and one 7901004 parts (four parts as opposed to the six on the card we have). These are both monolithic parts in 14-pin dual in line packages. They would not fit in a 10-lead TO-5 can. So this can’t be the same as the card we have.

What does this mean? It means the original revision of the card used hybrid devices. At some point, as the technology advanced, we created the 1000 series devices and the card was redesigned using the 1000 series to obtain the equivalent functionality. The card parts list revision E (dated 1969) uses the 1000 series devices. The computer system did not have earlier revision parts lists which would have shown the initial use of the hybrid devices back in 1963. So I could not determine when the changeover occurred.
So, we are both correct. CP-667 was originally designed using hybrid integrated circuits in cans. Some of the CP-667 hybrid devices in cans were also made by Fairchild and Texas Instruments. At some point, the cards were redesigned to use the new 7901003 and 7901004 devices. The first user of both the 7901000/7901001 flat pack version and the 7901003/7901004 dual-in-line version is listed as CP667. I wonder if there are actually CP-667 cards using the flat packs. The first program I know of that used the flat packs was UYK-7 and it used lots of them. {Editor's note: The 1824 used a 10-lead monolithic flat-pac circuit. The CP-901 used the 7901000 and 7901001 monolithic flat-pac components a year and a half before the UYK-7 design.}
We have no examples in our archives of the CP-667 27-pin cards using these newer monolithic devices. We do have another older CP-667 card with a couple cans on them but these are dual transistors in a 6-lead TO-5 can (part number 7900043) but I don’t believe these were ever made monolithic. By the way, the dual transistor has Motorola part number SM1000. Do not confuse this with the 7901000 part number for the integrated circuit. Motorola did not make the 7901000 device, at least not initially. The first suppliers for the 1000 series were Westinghouse and Raytheon.
By the way, it is interesting to note that on the early revisions of the 79003xx series drawings for the first monolithic devices which were used in the 1824 (MMRBM) computer (yes, these preceded the development of the 7901000 series) have a statement as follows: “fabricated on or in a single silicon chip, with deposited or diffused interconnect.” This obviously separates them from the earlier hybrid technologies. I guess we didn’t need to say this when we later generated the 1000 series specs since I doubt that a hybrid approach would have fit in those packages. [Larry Bolton}

From: Jorgen Andersen - I transferred a CS-1 compiler from the 1206 and constructed a FORTRAN IV compiler for the CP-667 during the late 60's.

From: Larye Parkins - I rediscovered the Legacy Site recently, noted an article about the CP-667, re no examples of the new flat pack cards that replaced the T05 cans. Don Steele and I overhauled the CP-667 in METC to upgrade it to the new cards. I somehow ended up with a failed one in my effects, which I still have, somewhere.

CP-667 Information:

Customer - BUSHIPS  Environment - Shipboard  First Delivery - 1964  Quantity built = 3
Vol/Ft3 = 62  Wt/Lbs = 2010  Pwr/Wts = 4200  Mem/Cyc = 2 usec  Mem/Cap = 131k core
I/O rate = 500 kw/sec  I/O channels - 16 duplex; -15v or -3v

George Gray is working on a CP-667 history paper - we'll provide info from that when it is published.

5. Repertoire Cards

Many people gave repertoire cards to the Legacy Committee. Keith Myhre scanned the cards before they were donated to the Charles Babbage Institute. The 36-bit cards are linked hereunder. There are copies of some of these cards at the Lawshe Memorial Museum. These are the CP-667 and early 1100 series; the 1107 through current 2200 series are linked from the Commercial Processors page.

6. Technical Manuals

The bit-savers web site ( has over 32,000 documents. We've copied the 36-bit ISA documents and linked them hereunder for technology researchers ease of access. Note that the 1107, 1108, etc computer series documents are linked from the Commercial Processors' page.


In this Chapter

  1. Introduction
  2. [left]
  3. Computers described: 2.1, 1103; 2.21105; and2.3, CP-667.
  4. Repertoire Cards
  5. Technical Manuals

We'd welcome inputs about any of these machines. [lab]

Chapter 56 edited 4/21/2022.