The 220 was perhaps the last major vacuum-tube computer design to reach the market. It did so as IBM, Philco, and Univac were announcing, and in some cases delivering, transistor designs. Because of this, the 220 did not sell particularly well, but it was popular with universities and research establishments, and it laid groundwork for the much more successful Burroughs systems that followed it, particularly the B5500.
This blog was re-titled for the Burroughs 205 and 220 a year ago, and we haven't said a peep about the 220 -- until now. This is Paul Kimpel with our initial post on that computer system. It will introduce the architecture and components of the 220, compare them to the earlier 205, and give a brief overview of our project to build an emulator and recover software for the machine.
Genesis of the Burroughs 220
By the mid-1950s, the ElectroData 205 had become a successful product. It was a relatively low-cost computer system for its day, but due to its drum memory and digit-sequential internal operation, it was also relatively slow. It had robust magnetic tape and punched-card facilities, however, and even though it had originally been designed with scientific computation in mind, these input-output facilities gave it an entry into businesses with a need to process large repositories of data -- the field of "commercial" data processing.
ElectroData needed to pursue these commercial customers and workloads if it was to stay in the computer business, and for that they would need a new, larger, faster system. They saw an opportunity to produce a larger system at a lower price point than competing systems of comparable size. The problem was that they were extremely strapped for capital just to continue their current operations, let alone to design and build a new system. The situation was so acute that, by early 1956, ElectroData was seriously considering terminating operations and closing the business. It was precisely at that point that Burroughs, having significantly more capital, and looking for an acquisition to boost their attempts at entry into the commercial computer business, came calling.
Prior to this, ElectroData had studied the possibility of modifying the 205 to use magnetic core memory instead of a drum. That idea proved not to be viable, for both technical and economic reasons, but it did initiate some preliminary work on a design, known as Project 737, for a new system that would use core memory.
As ElectroData's acquisition by Burroughs was closing, this new design was chosen over the BEAM IV, a system designed under the direction of Issac Auerbach at the Burroughs Research Center in Paoli, Pennsylvania. Project 737 became the Burroughs 220. While Paoli had their BEAM IV canceled, they did get to build the core memory for the 220.
The 205 and 220 are sometimes referred to as the B205 and B220. This is incorrect. The Burroughs B200-series systems, which eventually expanded to the B100/200/300/500-series, came immediately after the 220, ca. 1960. They were small, transistorized, variable field-length, character-addressable computers. Architecturally, they were very different from the 205 and 220. They were, however, somewhat similar to -- and competed against -- the IBM 1401.
Since the 220 is an outgrowth of the 205, and retained a number of the 205's characteristics, let us first review the architecture of the 205 and then see how the 220 differed.
A Review of the ElectroData 205
The 205 was initially conceived at Consolidated Electrodynamics Corporation (CEC) as a digital calculator. CEC made mass spectrometers, and analysis of the data coming from these instruments requires solving systems of simultaneous linear equations. By the mid-1940s, CEC had built analog devices that could solve systems with up to 12 unknowns, but to solve larger problems they needed to go digital.
CEC recognized early that the calculator would have additional potential if developed as a fully-programmable computer, so they expanded their scope and designed a system initially known as the CEC 30-201. The development proceeded through models -202 and -203, at which point CEC spun off the computer business as ElectoData, and the 30-203 became the Datatron 203. The 203 had only paper-tape input/output, plus a 10-character per second Flexowriter printer. Enhancements to support magnetic tape resulted in the Datatron 204, and finally adding the Cardatron punched-card interface resulted in the Datatron 205.
|ElectroData/Burroughs 205 System|
205 Drum Memory
The design of the 201 through 205 systems was oriented around their drum memory. Data was stored on the drum as 44-bit words, with each word organized as 11 four-bit BCD digits. The high order digit was used as the sign, with the low-order bit of that digit indicating positive (0) or negative (1). The other three bits of the sign digit were not generally defined for internal processor operations, but did have significance in some I/O operations. The remaining 10 digits of a word represented either a signed-magnitude decimal number or five alphanumeric characters. As an option, the later systems supported floating-point arithmetic, which used the high-order two digits of a word as an exponent biased by 50 and the low-order eight digits as a fractional mantissa. The sign digit retained its meaning in floating point as for integer values.
The memory drum was organized in 24 tracks of 200 words each. It rotated at 3750 RPM, so a full revolution took 16.8 milliseconds. Switching between tracks took negligible time, but the average latency until a desired word came under the read/write head was half a revolution, or 8.4ms. This was slow -- compare it to the IBM 650's drum, which rotated at 12,500 RPM. The choice of a slower drum was apparently made for reasons of cost and reliability.
|205 Memory Drum|
20 of the tracks on the drum formed the computer's main memory of 4000 words. To help compensate for the slow access time, the other four tracks were organized as recirculating delay lines, termed the high-speed loops, with the read and write heads separated by one-tenth of a revolution. This allowed storage of only one-tenth the number of words (20) per track, but with one-tenth the access time (0.84ms average). Special instructions could block 20 words in the main memory to or from one of the loops in a single drum rotation. Executing instructions and accessing data in the loops was almost ten times as fast as main memory. Doing so was highly desirable, but programming for it was complicated by the need to block and unblock words explicitly between the loops and main memory.
The drum was so central to the design of the system that even the clock signals for the electronics came from the it. Prerecorded tracks provided pulses for digit and word timing, plus a once-per-revolution signal to assure that the processor's address coincidence logic stayed in sync with the drum. These timing tracks gave the 205 an effective clock rate of 142.8 KHz.
205 Registers and Data Flow
The 205 had the following major registers:
- A -- the accumulator.
- R -- the "remainder" or accumulator extension of 10 digits. Digits could be shifted between this register and A, but R could not be loaded or stored directly. The Multiply instruction multiplied A by a value in memory, producing a 20-digit product in A and R. The Divide instruction divided the 20-digit value in A and R by a value in memory, leaving the quotient in A and the remainder in R. R was also used to store the return address in subroutine-branch instructions.
- B -- the index register. The ElectroData 203/204/205 were the first commercial systems in the U.S. to offer this form of address abstraction, although the idea appears to have come from the UK Manchester Mark 1 by way of Harry Huskey, who consulted with the original CEC design team.
- C -- the "command" register holding the current instruction. This 10-digit register held the two-digit operation code, the four-digit effective operand address, and the four-digit address of the next instruction to be executed. By default, the instruction address incremented automatically during fetch of an instruction into C. Branch instructions simply replaced this default address with the effective address from the branch instruction.
- D -- the "data" register. It buffered a word coming from memory or a peripheral device.
Since the data came into and left the processor in a digit-sequential manner, all internal processing was done digit-sequential as well. The adder was only one digit wide, and all inter-register transfers occurred one digit at a time. In fact, all inter-register transfers went through the adder, which would actually add or not, depending on the operation. The digit-sequential design also reduced the complexity of the electronics, helping keep the size and cost of the system down.
|Burroughs 205 Processor Data Flow|
205 Instruction Format
Words holding executable instructions had the following format:
± CCCB PP AAAA
- ± is the sign digit. If the low-order bit of this digit is 1 (negative), the contents of the B register are added to the AAAA field as it is transferred into the C register, thus indexing the operand address prior to instruction execution.
- CCC are the so-called "control" digits. These had significance primarily for I/O instructions.
- B is the breakpoint-skip digit. Bits in this digit would cause the computer to halt or skip the instruction based on switches set on the Control Console. It was used for program debugging.
- PP is the operation code, e.g., 74 for Add or 20 for Change [Branch] Unconditionally.
- AAAA is the operand address. In most instructions it referenced an address on the drum, but in some instructions, such as shifts, it signified a count, and in others it had no significance. B-register modification of this field would occur regardless of its significance.
The Burroughs 220 Architecture
Both architecturally and electronically, the 220 was quite similar to the 205. It was still a decimal, vacuum-tube design. The 220 retained the 205's 11-digit word format, instruction format, and numeric representation. Core memory replaced the drum. Memory sizes could range from 2,000 to 10,000 words in 1,000-word increments. A typical memory size was 5,000 words. There being no drum, the clock was supplied by an oscillator circuit, and at 200 KHz, was only slightly faster than the 205's 142.8 KHz rate.
The Operational Characteristics manual describes the architecture, instruction set, and input/output devices in detail.
While the 220 retained the instruction word format of the 205 and many instructions worked similarly, the operation codes and control (now termed "variant") digits were entirely different. Programs for the 205 had to be rewritten to run on the 220. That said, the 220 was a much easier system to program than the 205, first because the design headache of blocking 20 words at a time into the high-speed loops was no longer an issue, and second because the 220 had a more powerful instruction set.
|Burroughs 220 System|
220 Registers and Data Flow
The 220 retained the same major registers as the 205, but with a few changes and additions:
- The A and D registers were the same, except that now data both entered and left the processor through D.
- The R register had the same function in arithmetic and shift operations, but was expanded to 11 digits to accommodate a sign digit. It no longer stored the return address for a subroutine call. R could also be loaded and stored directly from and to memory, and could be compared directly to a word in memory.
- The C register was still 10 digits and held the op code and effective address, but not the address of the next instruction to be executed (see P below). It was a copy of the instruction word from memory, without the sign, and after any B-register indexing of the address field had been applied. Therefore, it also held the variant digits from the instruction word.
- The B register was the same, although there were additional instructions for loading, incrementing, decrementing, and branching based on its value.
- A new register, P, was implemented to hold the next-instruction address.
- A new register, E, was implemented to hold the address of a word to be loaded from or stored to the core memory.
- A new register, IB (input buffer), was implemented to hold the word loaded from or stored to memory. Words were generally transferred to and from this register and the D register, but not always.
- A new register, S, was implemented to replace the function of the breakpoint digit in instruction words. You could set an address in this register and enable it with switches on the console. When execution reached that address, the processor would halt.
|220 Control Console|
The 44 bits of a word were transferred in parallel between the core memory and the IB register in a single five-microsecond clock cycle. Then the data moved between IB and D, and through the other registers, in the same digit-sequential fashion as in the 205. The adder was still only one digit wide, and most inter-register movement continued to be gated through it.
|220 Processor Data Flow|
This sequential operation and the only modestly-faster clock meant that the internal performance of the 220 was little better than that of the 205. Most of the system's speed increase came from eliminating the latency of the drum memory.
Retention of digit-sequential operation appears to have been a conscious decision, probably to reduce the cost of the system and allow it to come to market more quickly. An internal ElectroData memo from 5 June 1956, after the Burroughs acquisition was announced but before it closed, stated:
The basic configuration of the 220 will be the same as the 204, with the exception of the Core Memory. New circuitry in the 220 will be held to a minimum. Generally the plug-in [vacuum tube] packages will remain the same.
We do know that the concept of the 220 was not fully realized in the design that was finally built, at least in the mind of its chief logician, Edward (Ted) Glaser. In his 1973 oral biography, on file at the Smithsonian National Museum of American History, he had this to say (page 24):
It turns out that what the 220 was supposed to be was the 205, with the loops removed, the recirculating loops, and a small core memory put in. That was my assignment. And it became quite clear that with the speed of core that was then available, this couldn't be done and it was a ridiculous balance. There was no way I could prove that to anyone. So I did one of the lowest-down, dirtiest tricks I've ever done. I designed five machines before I told anybody about any of them. And each one had glaring faults. And I was able to escalate Burroughs from one machine to the next. Well, it turned out we got stuck on four. We didn't build five. So four -- 220 was the fourth in the series. It was not the machine I really wanted to build, but it was close to it.At that point, Glaser appeared to start to talk about the fifth design, but the interview was interrupted, and it never came back to this topic. So we'll never know what he had in mind for his fifth design, or the ways in which he thought the 220 as built was still deficient.
220 Partial-Word Operations
Aside from the advantages that retaining the digit-sequential operation had for cost and time to market, the most beneficial was that it enabled a very useful new feature -- operation on partial words.
The digits in a 220 word are numbered as follows:
± 1 2 3 4 5 6 7 8 9 0
where ± is the sign digit and "0" is interpreted as 10. A "partial word" is a contiguous field of digits denoted by a two-digit pair, referred to as sL (start-Length). The first digit of the pair specifies the starting digit in the field; the second digit specifies the field length. The field therefore consists of the digit numbered s and the L-1 digits to its left. A length of zero is also interpreted as 10.
Thus, the operation code in a word can be referred to as the 62-field -- digit #6 and the digit to its left. The low-order four digits comprising the operand address are the 04-field. The entire numeric portion of the word (all but the sign) is the 00-field. The sign could not be the starting digit of a field, but in most contexts could be included in a partial-word field. Field overflow was not permitted, however -- the sign digit was as far to the left as a field could go. If an sL digit pair had L > s+1, the processor would halt with a Program Check alarm. An sL = 12 field is valid; a 13 field is not.
Partial-word operations were implemented as variations on shifting. Like the 205, the 220 could physically shift or rotate its registers only to the right. Left shifts were actually rotates to the right for the complementary number of digits. In a partial-word operation, the word is simply rotated to the right until the s digit is in the low-order or "0" position, from which it can be transferred to and from the adder. Then it and the next L digits are operated upon (add, transfer, etc.), and finally any remaining digits are again simply rotated until the word is back in its original position. This behavior was mechanized using a special register, DC, the Digit Counter.
Partial-word fields were supported in the following operations:
- Branch on equality of a two-digit value compared circularly to a field in register A or R.
- Compare a field in A or R to a corresponding field in a memory word and set the Compare Indicator, from which branches could be made based on a low, equal, or high conditions.
- Increase or decrease a field in a memory word by a four-digit value, leaving the other digits in the word unaffected.
- Decrease a field in a memory word by a four-digit value (possibly zero), then load the high-order digits of the resulting value in the field into the high-order digits of B.
- Store a field from A, R, or B to a corresponding field in a memory word, leaving the other digits in the memory word unaffected.
The partial-word facilities also provided convenient and efficient ways modify operand addresses, variant digits, and even op codes in instruction words. A common trick was to implement a switch by adding or subtracting 29 to the 62-field in a word -- toggling the op code between 01 (No-Op) and 30 (Branch Unconditional). Partial-word facilities could also be used to pack multiple values into a single word, allowing programs to maintain larger numbers of values -- a not-insignificant advantage in the days of severely-constrained memory sizes. This was particularly important in the implementation of the Burroughs Algebraic Compiler (BALGOL), which will be discussed in future blog posts.
Additional 220 Enhancements
There were several other notable enhancements to the 220 instruction set over the 205:
- The A, R, and B registers could be cleared to zero individually or in any combination with one instruction. The 205 could clear only A and R, and A only as a side effect of a store.
- R could be loaded and stored directly without affecting A. The 205 could load and store R only by shifting values between A and R. This improved the setup for integer multiply/divide operations and allowed R to be used more conveniently as temporary storage. Stores could be of a partial word.
- Similarly B could be loaded and stored directly, either as a whole or partial word.
- A word in memory could be cleared to zero with a single instruction.
- Instructions to compare A and R to a whole or partial word in memory replaced the 205's awkward branch-on-sign-difference mechanism.
- The sign digit of the A register could be tested against or replaced by a literal digit value.
- Loop control was considerably enhanced. The 220 supported the 205's ability to decrement B and branch if it had not underflowed below 0000, but the decrement value could be 0000-9999 instead of just 1. Similarly, there was an instruction to increment B by 0000-9999 and branch if the result had not overflowed past 9999.
- Looping could also be controlled using a new element, the Repeat Toggle. This was set by instructions that could increment or decrement a partial-word field in a memory word by 00-99. The toggle was set if the operation did not overflow or underflow, respectively, the designated field. The Branch Repeat instruction would branch or not based on the state of this toggle.
- Subroutine call was supported on the 220 by the Store P instruction, which would store the value of the instruction counter, plus one, into the low-order four digits of the word at the instruction's operand address. This effectively stored a return address, and would typically be followed by a branch instruction to enter the subroutine. Compare this to the 205's Change-and-Record instructions, which stored the return address in the upper four digits of the R register.
- The 220 supported a full complement of shift operations. Right shifts discarded digits off the low end of the register; left shifts acted as rotates (and as mentioned above, were implemented internally as right rotates). Register A could be shifted left or right, with or without participation by the sign digits. A and R could be shifted together left or right for up to 20 digits. The sign digit did not participate in the twin-register shifts, but it did get copied from A to R or R to A, depending on the direction of the shift.
- The 220 had a Record Transfer instruction that could move from one to 100 words between locations in memory. The instruction's operand address specified the starting source location; the B register specified the destination address. At the end, B had been incremented to the address after the last word transferred, allowing multiple consecutive transfers to move larger blocks of data.
- On the 205, overflow in an arithmetic instruction caused the processor to halt if the next instruction was not a conditional branch. On the 220, this behavior could be turned on or off programmatically. There was also an explicit Branch on Overflow instruction to test for overflow and reset the indicator if the branch was taken.
- The 220 had an explicit No-Op (no operation) instruction that did nothing. On the 205, any unassigned op-code was treated as a No-Op. On the 220, executing an unassigned op-code caused the processor to halt with a Program Check Alarm.
220 Input-Output Facilities
The 220 supported the same types of input-output devices as the 205, although the I/O instructions and the capabilities of the devices in some cases were quite different.
The so-called console devices included teletype printers, paper-tape readers, and paper-tape punches. A system could support up to 10 readers, up to ten teletypes or punches in any combination, and one special teletype termed the Supervisory Printer, or SPO (pronounced "spoh"). Teletype units could be substituted for punches and vice versa.
The 205's Frieden Flexowriter was replaced by a Teletype Model 28 RO (Receive Only) teleprinter, which operated at 10 characters per second. The Model 28 had the advantage that it used a moving type box to position characters on a line rather than the Flexowriter's moving typewriter carriage. The SPO was identical to the other teletype devices and was interchangeable with them. The SPO was addressed by a different instruction in the processor, however, and that instruction had an automatic decimal-point insertion feature that instructions for the other teletypes did not.
ElectroData designed and built their own high-speed paper-tape equipment, using 7/8-inch (22mm) tape and a seven-level code compatible with the IBM Eight-Channel Tape Code for the Type 046 Tape to Card Converter (a variant of the 026 keypunch). The reader was photoelectric and operated at a switch-selectable 500 or 1000 characters per second. It could stop in the space of a single character. The punch operated at 60 characters per second.
As an option, each teletype device could be configured with a paper tape reader, but this could be used only for off-line printing of paper tapes.
The paper-tape punches, teletypes, and SPO were connected to the processor in a daisy-chain. Switches in the device cabinets designated their unit number, 1-10, or SPO. The first device in the chain with a designation that matched that of an output instruction received and acted upon the data. The teletypes and SPO could each respond to multiple unit numbers, so it was possible for the software to address multiple devices but have the output come out on a smaller number of them.
On the 205, words were output to console devices one at a time from the A register. They were formatted as numeric or alphanumeric based on the values of address digits in the instruction words. On the 220, a single instruction could output up to 100 words directly from memory. The 2-bit in the sign digit of a word indicated whether it should be formatted as numeric (0) or alphanumeric (1). Thus, words with signs of 2 were considered to hold five alphanumeric characters, and words of mixed numeric and alphanumeric data could be output by the same instruction. This alpha vs. numeric behavior was effective only for the console devices, however.
Punched Card Devices
The Cardatron punched-card subsystem originally developed for the 205 was adapted for the 220 with very little change. The Cardatron was an interface between the 220 and IBM punched-card equipment. It translated between card or printer columns on the IBM machines and words in the 220 memory, It provided low-level editing of the data during transfer by means of digit strings termed "format bands" that were loaded to the interface from the 220's memory. It could also buffer one card or line image per IBM machine, allowing the processor to continue running while the machine operated on that card or line.
A 220 could have one Cardatron subsystem, which could support up to seven IBM machines. The Type 087 collator was used as a 240-card/minute reader, the Type 523 could be used as a 100-card/minute reader or punch (but not both at the same time), and the Type 407 was used as a 150-line/minute printer.
The only significant differences between the Cardatron for the 205 and the 220 were (a) the ability to specify a sign of 2 for a word during input translation from a card reader, and (b) data transfer between the 220 memory and Cardatron units went from high to low addresses. That is, Cardatron instructions specified the ending address of the buffer area in memory. The Cardatron processed card and print columns backwards, so the descending addressing allowed the data in memory to be in ascending order. On the 205, data transfer was in ascending sequence, which meant that the words in memory were reversed with respect to their order on the card or print line.
For more information on the Cardatron and its features, see these two posts on this blog for the 205 version:
Magnetic Tape Devices
The 220 took the basic idea of the 205's magnetic tape subsystem and significantly extended it. The 205 used 3/4-inch (19mm) plastic tape on 2500-foot (762m) reels. The tape had two parallel lanes that each stored digit-wide frames at 100 bits/inch and moved at 60 inches/second. Only one lane could be read or written at a time, but the lane was programmatically selectable. Data was recorded in fixed-length blocks of 20 words each. Individual blocks could be overwritten in place. A full reel of tape could hold 10,000 blocks. Each block was assigned a four-digit number. The tape subsystem could search bidirectionally for a specified block number independently and asynchronously from the processor.
The 220 enhanced the 205 tape implementation in the following ways:
- The tape was physically identical, but came on 3500-foot (1067m) reels.
- The tape still had two parallel lanes, but data was recorded at 208.33 bits/inch and the tape moved at 120 inches/second -- twice as dense and twice as fast.
- Blocks could be 10-100 words in length, and a tape could have mixed block sizes. Blocks could still be overwritten in place, but once a block of a certain size was written on tape, that size could not be changed. The entire tape had to be erased (a process known as "editing") and rewritten in order to change block sizes.
- Up to ten blocks could be read or written in a single instruction. Variants of the instructions allowed reading and writing sequences of mixed block sizes. Tapes could also be spaced forward or backward up to ten blocks at a time.
- Instead of block numbers, the first word in each block was considered to be its identifier or "keyword". The tape subsystem could search a tape bidirectionally and asynchronously for the block with the highest keyword that was less than or equal to a specified value. Optionally, the match could be made on a partial-word field. The search stopped with the tape positioned to read the matching block. In order for the search to work as intended, the keywords had to be unique and written in ascending order, but it was not a requirement that tapes be created in that particular fashion.
- In addition to the bidirectional keyword search, the tape subsystem could search asynchronously in the forward direction for a block where a specified word in the first ten of the block matched a value. This was termed a "scan." This match could also be made on a partial-word field. As with search, a scan stopped with the tape positioned to read the matching block.
- Instead of a simple end-of-file mark like the IBM tapes had, the 220 could write and sense two types of special blocks on the tape:
- An End-of-Tape block was typically written at the end of information on a tape, but could be placed anywhere. It contained a single control word (although the block physically occupied ten words on tape) that held two addresses:
- The 64-field of that control word was the address of a word in memory where the values of the P register and 04-field of the C register would be stored.
- The 04-field of the control word was the address to which the processor would branch when the block was read.
- A Control Block was one where the sign digit of its keyword was 7. When this type of block was read, all words in the block except the last were transferred to memory. The last word was not transferred to memory, but was acted on in the same way as the control word in an EOT block, storing fields from the P and C registers and branching to a specified address.
|220 Free-standing Tape Drives|
|220 DataFile Tape Unit|
This image shows the interior of a 205 DataFile, but the internal mechanism for the 220 was similar:
|205 DataFile Showing Tape Strips|
Burroughs developed an early electrostatic printer named the Whippet that could output about 250 characters per second. It used coated paper and heat to fix a dry ink to the surface of the paper. It was originally developed for the U.S. Army Signal Corps, but at least a few units were installed at non-government sites, including one at CalTech. It could substitute for one of the teletypes.
Burroughs also developed its own high-speed drum printer for the 220, a design that continued to be refined and used into the 1970s with the B-series computer systems. The printer for the 220 could be used in two ways. It could be attached directly to the system and driven programmatically. It could also be used off-line, with a separate control unit that read data from a free-standing tape drive. The control unit was configured through a plugboard and was designed with the intent that data could be formatted and printed directly from master files, avoiding tying up the rest of the system for long print runs. The idea of off-line printing did not survive beyond the 220 era, however.
The retro-220 Project
Like my earlier emulation projects, the retro-220 emulator is open source and available for anyone to use. The emulator itself and all but the magnetic tape implementation were substantially complete by late Spring 2017. The 220 tape subsystem proved to be a significant challenge, and was not completed until late 2017. At this writing, the emulator is at version 0.06, and is complete except for the DataFile and high-speed drum printer devices.
This screen shot shows the emulator in action:
|retro-220 Emulator Running from the Firefox Browser|
The second goal for this project has been reconstruction of the Burroughs Algebraic Compiler for the 220, better known as BALGOL. This was an early implementation of the Algol-58 language, the predecessor to the better-known Algol-60. The compiler was first released in March 1960. Donald Knuth, who participated in its development, has donated a scan of the compiler and library listing to the Computer History Museum in Mountain View, California. An earlier, less complete listing is also available on bitsavers.org.
I have used Professor Knuth's listing to transcribe the assembly language source code for the compiler and its library, and have written two cross-assemblers to generate 220 object code from those transcriptions. The reconstructed compiler is now starting to work and is presently undergoing further testing and validation. The current state of its development can be found in the retro-220 project repository.
At this point, I am starting to work on a wiki for the emulator and compiler that will be available on the repository web site. I am also preparing additional blog posts on the project. The next one will be an introduction to the emulator and its user interface, so stay tuned.