Why were chips socketed in early computers?

Question

In many early computers, many of the chips were in sockets rather than soldered directly to boards, e.g. this series of pictures of the Tandy CoCo 1 has a note to the effect that all the chips are socketed: http://tech.markoverholser.com/?q=node/50

I can understand doing this with RAM chips, because it's often feasible and desirable to upgrade these; I think the CoCo 1 did often receive RAM upgrades.

But what was the purpose of doing it with all the others? I would expect soldering all the other chips directly to the board, would both save money and improve reliability by eliminating one thing that can go wrong.

Answer 1

Caveat: It might be useful to distinguish between high volume low cost computers (like the mentioned CoCo) and low volume high cost machines (like Intel boards - or workstations). I assume the question to be rather about these high volume low cost machines.

---

Why were chips socketed in early computers?

There are several reasons:

• Most important: Chips where not fitted by robots back then, but by hand. Thus any chip fitted wrong and soldered would make the whole board unusable (*1), while with sockets it's a simple swap to make it work.

• Second, sockets can endure more abuse (in this case heat) than chips, thus having them inserted later increases acceptable tolerances during manufacturing and leads to a lower rate of failure (*2)

• Next, fitting sockets instead of chips decouples production. Boards can be prepared, ready made and stocked until the time there are orders to fill.

• As a result investment can be reduced by ordering the most expensive single part(s) -- the chips -- as late as possible (*3).

• Similarly, this is helpful to keep board production running even when some chips are not available for some time (*4).

I would expect soldering all the other chips directly to the board, would both save money and improve reliability by eliminating one thing that can go wrong.

Yes. It does, as soon as there are automated placement machines and -- more importantly -- an order of sufficient size to keep capital cost in check.

I can understand doing this with RAM chips, because it's often feasible and desirable to upgrade these; I think the CoCo 1 did often receive RAM upgrades.

Not really, as that's something users may want, not necessarily the manufacturer.

---

As said before, this is focused on high volume/low cost machines. While all of the reasons are as well valid for the high performance/high cost boards, labor and stock cost are somewhat less important due to labor which is already rather intensive for small production runs, and stock cost aren't that big in the first place.

Further, the repair aspect wasn't any consideration for cheap, mass produced machines. Commodores share of a C64 was less than 150 USD during most of the time (in the late 1980s even less than 50 USD). There is not enough money to be made in repairing defective units beyond a bare function check. Even more so with the lower priced CoCo. So serviceability wasn't a major concern.

---

*1 - Yes, they could have had people assigned to desolder the chip(s), clean the through-holes and have new ones fitted and soldered. Except, that would have cost possibly more than the wholesale price of that (cheap) computer would have been.

*2 - Wave soldering is a rather secure process and around since at least 1947 (oldest usage I know). It does require to have the whole board plus all of the placed components preheated to 130-160 degree Celsius (depending on the process) for several minutes. In combination with (somewhat) early chips, this may cause failure within the chips. So using sockets instead during soldering was a great idea.

*3 - One reason why Commodore continued to use sockets for the more expensive chips even after using automated placement.

*4 - Commodore is well know to have run into these conditions several time - for example with some batches of C16.

Answer 2

One reason to use sockets at the beginning of a production cycle is to make it easier on the service technicians.

At one company making computer terminals, the techs would identify a bad chip and then give the board to the rework person to replace. This destroyed the presumed-bad chip and risked damage to the board. Before the techs were familiar with the board signals and failure modes, the false-bad rate would be pretty high. With experience, the false-bad rate got really low, and all boards were built without sockets on all parts except the ROMs, which could be upgraded in the field.

When times got busy and the bad boards piled up (the dog pile), some of us from engineering would volunteer to go to the factory and help fix boards. I was really good at debugging design issues, but the techs were much faster than I was at debugging production problems. I learned a lot from working with them.

Answer 3

There was a time when many personal computer manufacturers did almost no incoming QA on components. Sockets allowed swapping out chips when the assembled computer didn’t run, e.g. didn’t pass initial power on testing or burn-in. Simply swapping out certain chips by trial-and-error was a very common debugging technique to find the bad ones, much harder to do if the chips are all soldered down.

Some personal computer models would not run with all worst case timing spec parts. Swapping some parts was likely to mix in some parts not as slow and allow the system to meet the timing necessary to run.

Socketing is no longer done today, as chip yields are higher, more testing and incoming inspection is done, sophisticated CAE tools allow far more complete worst-case timing analysis before a product goes into production, and modern automatic test equipment allows automatic pinpointing of many common faults.

Answer 4

I suspect this is down to quality engineering (QE) — in the early micro days many chips were susceptible to failures (during manufacturing) but wouldn’t be as well-tested as they are now, so the chances of getting bad batches were higher. If you finish assembling a micro and discover that one of its chips is bad, it’s easier to fix if it’s socketed.

Another aspect is that it’s easier to solder a socket than a component, or rather it’s harder to damage a socket when soldering it than it is to damage a component. That might not be relevant when manufacturing at scale though (I’m not sure when wave soldering became common).

Answer 5

A long time ago you could solder an entire Acorn Atom yourself. A hobbyist who was unsure about his soldering skills would use sockets, thinking that he was playing it safe. I helped diagnose a non-working board where the pin for the CPU clock made no contact with the socket.. The signal on the bottom of the board was good, it took us hours to find this. The customer had used cheap sockets with contacts only on the inside, on all ICs. Bending all pins inward before pushing the ICs in the sockets fixed that. You could even buy special DIP IC inserters.

Tin plated ICs with gold plated sockets (Garry/Augat) don't go well together, the gold dissolves in the tin and ultimately all contact is lost.

The added height due to the sockets increases the inductance of the tracks a lot, leading to bad HF behavior and bad EMC. This was OK when computers were slower and there were no FCC regulations, but not now.

All reasons why the CE industry uses no IC sockets anymore, except maybe for EPROMs which must be removed for erasing and programming. And for CPUs where the customer makes the choice, but then the socket is an expensive one.

Answer 6

I am not to what extent this was a manufacturer consideration, however it made the machines much easier/cheaper to service. This meant customers would be more willing to "invest" what might be $1-2000 in today's money since they could continue using it for a longer period.

As testament to this, having never touched an 8-bit machine before, I recently got my hands on two dead C64s and within twenty minutes I got one working by swapping chips.

Certainly in other manufacturing sectors, serviceability is a consideration so I find it likely the manufacturers did consider this.

Answer 7

One area that use to mandate the use of sockets (and still does to my knowledge) is in the area of huge-footprint chips (specialized processors, CCD arrays above a certain size, Array-Processor Logic for in-circuit AI boards, etc.) where the mechanical stresses of insertion and/or pick-and-place soldering across the face of the chip exceed a certain recommended threshold.

In such cases, they're likely to use a ZIF-socket (Zero-Insertion-Force) where these high-dollar frequently upgraded components can be removed as needed, particularly in environments where ambient heat, vibration stresses, and radiation may cause gate degradation. It still happens, and when your board costs $1.75 and your AI Gate-Array chip costs $20K, you'd prefer to protect your investment.

The other variable not addressed earlier is when chip failures are not entirely understood and diagnostics on specialty high-density chips differ between the actual (working) environment and the prefab test environment. Soldering in these huge beasts (even when possible) usually means that the chip is destroyed during the desoldering process and therefore means it is no longer possible to analyze the casualties for failure modes after-the-fact.

Answer 8

It's really quite simple; early computer MBs were designed to accept a number of different processors. One could easily 'upgrade' their processor by simply swapping in a more powerful one.