Is it good design to put SMT components beneath socketed DIP packages?

[technix]

For example, a 68010 on a mostly SMT board, can I socket the DIP-64 68010 and put the following items beneath it:
* Bus pull-up resistors, 0603x4 resistor packs,
* Bypass caps, 0603 ceramic caps
* Reset circuitry, SOT23-4 chips, SOD123F diodes and 0603 resistors
* Part of the level shift circuitry, a 74HC00 in SO-14 package and a 0603 resistor.

Even more daring, can I squeeze a DC-DC 12V-5V converter down there without interfering the chip? (TPS562200 in SOT23-6 + 0603 passives + EIA case B tantalum cap + a low profile power inductor.)

[helius]

It was common to place bypass capacitors underneath DIP chips (and there were special variants of DIP sockets that contained pre-installed leaded capacitors between VCC and GND). Some of the other ideas can work, but the magnetics may cause EMI problems.

[bloguetronica]

Hi,

IMHO, although it is not advisable to do so, it is perfectly doable. The reason behind this is that the socket will make the replacement of these components much more difficult. But if these SMD components are all passives (resistors, capacitors, etc), the chance of having to replace them is very slim, and it is perfectly OK to do so. Integrated circuits and oscillators should be placed elsewhere.

Kind regards, Samuel Lourenço

[glarsson]

The placement of the components depends on the socket. All sockets are not equal, so changing socket model might require PCB layout changes.

[bson]

You can also put all SMD parts that make sense on the other side, wherever is convenient after layout of the TH parts on the board.  The parts that wouldn't make sense would be things like indicator LEDs.  With corner holes and standoffs (the standard issue M3 stuff used to assemble PCs) it won't sit on the SMD parts when you place it on a table.  (Or the TH pins for that matter.)

[bson]

BTW, skip the resistor arrays, they make little sense in SMD.  Just use discrete resistors, even placed side by side they take about the same space, but you also have the option of placing them wherever you wish, staggering, etc, so you get more routing flexibility by just sticking with plain discretes.

[T3sl4co1l]

BTW, skip the resistor arrays, they make little sense in SMD.  Just use discrete resistors, even placed side by side they take about the same space, but you also have the option of placing them wherever you wish, staggering, etc, so you get more routing flexibility by just sticking with plain discretes.

I disagree, at least unless you're going to use 0201s.  0402s are the same size as a 1206 quad.  You do get more flexibility on where to place the individuals, if you need to fill in an oddly shaped space, of course.

Arrays also reduce BOM size and assembly cost.

I ADORE them for hand assembly.  Four resistors for the effort of one!  Less advantageous for PnP but still saves time.

You can also consider using a single array for just one resistor at a time, if you discover you need certain lucky values:
https://www.edn.com/design/components-and-packaging/4421194/Resistor-combinations--How-many-values-using-1kohm-resistors--
from a 1k, you get: 4k, 2.5k, 1.67k, 1.33k, 1.00k, 750, 600, 400 and 250.

This gets more difficult to spin values if you find you need a different particular resistance, of course.  A better option for cost reduction going into production, than for prototyping.

Regarding putting stuff around a socket: sure, resistors and capacitors are fair game.  PITA to rework, be very sure you're only placing components you know will never need to change!  I wouldn't put a power converter there -- but, I wouldn't balk at putting it on the opposite side, on a 4 layer build where the inner planes provide shielding between the converter and CPU.   (2 layer, nah.  But I would guess if you're pushing for higher density, you're more than willing to pay for 4 layers.  You may also want to look up QFPs or PLCCs of those processors -- they're still around, many of them!)

Tim

[langwadt]

Hi,

IMHO, although it is not advisable to do so, it is perfectly doable. The reason behind this is that the socket will make the replacement of these components much more difficult. But if these SMD components are all passives (resistors, capacitors, etc), the chance of having to replace them is very slim, and it is perfectly OK to do so. Integrated circuits and oscillators should be placed elsewhere.

Kind regards, Samuel Lourenço

put the socket on the reverse side of the board and many of those issues go away

[technix]

Hi,

IMHO, although it is not advisable to do so, it is perfectly doable. The reason behind this is that the socket will make the replacement of these components much more difficult. But if these SMD components are all passives (resistors, capacitors, etc), the chance of having to replace them is very slim, and it is perfectly OK to do so. Integrated circuits and oscillators should be placed elsewhere.

Kind regards, Samuel Lourenço

put the socket on the reverse side of the board and many of those issues go away

This beats the point of running tucking SMT component beneath socketed THT chips: I get to run up to four layers of components this way: socketed THT chips at the layer above component side, SMT at component side, SMT at solder side, and socketed THT chips at the layer below solder side.

[hermit]

I did this for a friend using a Teensy but we used header pins instead of a socket so that gave us lots more space.  Since he was't using many pins on the Teensy we didn't add all of the pins either.  I think one or two for extra stability on the end.

[Benta]

Technix, you should take look at the old PEP Modular Computers' half-height VME boards. They used elevated sockets for the 68010 and placed DIP parts (bus buffers etc.) as well as decoupling caps underneath. Same thing for the peripheral/memory chips, they'd be raised and the PCB real estate underneath used.
Anything goes if you're space limited. Just do it.

[djQUAN]

Might not be an issue but a possibility, the stress of inserting or removing a socketed chip could cause breakage of SMT components nearby and underneath especially long/big ones like the array or power inductor.

[bson]

I disagree, at least unless you're going to use 0201s.  0402s are the same size as a 1206 quad.  You do get more flexibility on where to place the individuals, if you need to fill in an oddly shaped space, of course.

Well, I'm looking at a row of 15 0402 resistors.  If I were to replace it with say a Vishay VSSR2403101JUF (QSOP-24, 0.64mm pitch) then it's narrower, about half the width, but the QSOP obviously has two rows of pads, plus 6mm of package or so on top of that.  Maybe I'm looking at the wrong parts?  A bunch of quads?  https://www.mouser.com/datasheet/2/427/vtsrvssr-88387.pdf

Here's what I'm looking at (ignore the silkscreen on vias):


Here's a QSOP-20 dropped on top for scale:


So, yes, it would fit in this particular case.  From a cost perspective, the 10x VSSR costs just over $1 qty 1k.  The Susumu RR resistors cost... $0.008 qty 10k (since it requires 10x as many).  Unless there's something to be saved during P&P I don't really see the economy here.  How can you compete with resistors that cost under a cent?  https://www.mouser.com/ProductDetail/Susumu/RR0510P-101-D?qs=sGAEpiMZZMvdGkrng054t%2fXDkuzyg5Mcu5v70Lq2dvM%3d  So $1000 vs $80?

Edit: I do like the idea of being able to route on the top copper layer under the larger array package.  That can help in tight spaces as it effectively adds another layer.  I'll have to keep that in mind...  Of course the assignment to an array would be a choice made during layout, not during initial design (capture).

[station240]

I wouldn't put a DC-DC converter under a CPU core, as it's just extra heat to get into the CPU.

[technix]

I disagree, at least unless you're going to use 0201s.  0402s are the same size as a 1206 quad.  You do get more flexibility on where to place the individuals, if you need to fill in an oddly shaped space, of course.

Well, I'm looking at a row of 15 0402 resistors.  If I were to replace it with say a Vishay VSSR2403101JUF (QSOP-24, 0.64mm pitch) then it's narrower, about half the width, but the QSOP obviously has two rows of pads, plus 6mm of package or so on top of that.  Maybe I'm looking at the wrong parts?  A bunch of quads?  https://www.mouser.com/datasheet/2/427/vtsrvssr-88387.pdf

Here's what I'm looking at (ignore the silkscreen on vias):


Here's a QSOP-20 dropped on top for scale:


So, yes, it would fit in this particular case.  From a cost perspective, the 10x VSSR costs just over $1 qty 1k.  The Susumu RR resistors cost... $0.008 qty 10k (since it requires 10x as many).  Unless there's something to be saved during P&P I don't really see the economy here.  How can you compete with resistors that cost under a cent?  https://www.mouser.com/ProductDetail/Susumu/RR0510P-101-D?qs=sGAEpiMZZMvdGkrng054t%2fXDkuzyg5Mcu5v70Lq2dvM%3d  So $1000 vs $80?

Edit: I do like the idea of being able to route on the top copper layer under the larger array package.  That can help in tight spaces as it effectively adds another layer.  I'll have to keep that in mind...  Of course the assignment to an array would be a choice made during layout, not during initial design (capture).

I am not using resistor packs for high precision purposes - just bus pull-up/pull-down, high speed signal termination and LED current limiting. No point getting that Vishal part, just those UniOhm ones at best.

[T3sl4co1l]

Like I said, quads are good.  That SO-ish plastic array, looks old fashioned or boutique!

For whatever reason, duals and quads are almost always a better deal than quads or octals or what have you.  Transistor arrays for example, duals are cheap and plentiful, quads are boutique.  Resistor arrays, you'll find ceramic quads and octals without much added cost, but the quads are probably the best deal overall.  Typically fractional-cent in thousands, I would guess they're basically no more effort to manufacture than regular 1206s.



As you can tell from the courtyards overlapping, I would never place 0402s quite this close together, but you could if you wanted to.  That's an, uh, Yageo YC164 series I think?  0.8mm pitch.

Tim