The Ware for July 2013 is shown below.
Ahhh, the 80’s….back when digital chip designers were actually circuit designers, and not coders; and back when you could actually see a transistor with an optical microscope. I don’t, however, miss the jean jackets, shoulder pads and leg warmers.
This fine ware was contributed by A.E. (full image credits to be given with the solution). Wonderful job on the microscopy!
First thought coming to me with the 80’s clue : Z80 processor
It’s going to be a bit more than a Z80 with 144 pins…
Could it be an Intel 80386? They were introduced in 1985 and came among others in a SQFP-144 package.
http://en.wikipedia.org/wiki/Intel_80386
Don’t think that it’s a 80386. Look at the bond wires, they seem to go up straight. Maybe I am mistaken, but in the 80386 they are connected in two layers:
http://www.cpu-galaxy.at/article/opening_die_core-Dateien/80386.JPG
I don’t have the book about it any more so I can’t check, but: the CPU for the Apple GS? The one with the built-in Apple ][? The last hand-drawn CPU (as it was claimed at the time)?
Apple II GS just an upgraded 6502 so still too small. I was thinking about a 386 or so, but the die pictures I found aren’t similar.
I think not it is Intel. Intel layout is distinctive.
Maybe a Motorola?
Looking at the die shot: My suspicion is that the upper-left rectangular, regular part is the register file. From counting the rows and columns, I would suspect this to have 32 registers with 32 bit each… (but I might have counted the wrong cell-size and maybe it’s only 16?)
The rectangular part below might be the ALU: 32 almost identical columns (bits), but vertically with non-repeating patterns (different ALU logical functions). The ALU looks pretty simple, so that would match a more RISC-type chip (that doesn’t do everything and the kitchen sink in hardware).
From Die-shots I’ve found online, ARM1 (which is 3 or 3,5µm of structure width) looks like it would be a little courser, but the structures look remarkable similar to what I called Register-File and ALU.
ARM3 has 1,5µm of structure width and already 4KB cache (late 80s)…
So I’d guess for a 32-bit processor with 32 registers, in about 2µm structure width. And maybe a few kB of cache? (the highly regular structures on the middle right).
Die-shots of some other processors linked from http://www.cpu-world.com/forum/viewtopic.php?p=168818 :
ARM1: http://www.flickr.com/photos/duboulaybiggs/5470486537/sizes/l/in/photostream/
ARM3: http://www.altparty.org/2009/tl_files/special_guests/sophie/arm3_board.jpg
I think it is a 32 bit device – although I see two portions that might be 32 bit data buses: lower and lower left edge as well as top left and left top edge. Then there is something like a 21 Bit bus on the right top and top right edge.
Could this be some kind of memory or bus controller? (Looked for PCI host controllers, but did not find any die photos…)
First, it roughly looks like a processor. And with that many pins, in the 1980’s, that probably means 32 bit.
But, taking a different direction, there aren’t too many 144 pin processors. And, with the 1980’s limit, there are VERY few.
I’d suggest a Freescale MC68302, 144 pin QFP.
it’s definitely a 32-bit ALU, hardware multiplier, 32 registers
it’s older – looks like 2 layers of metal – so 90s?
harvard architecture, so probably a DSP
256×32 sram (8k bits)
looks like a dac (or adc?), in the top right hand corner
and maybe a VGA in the middle of the RHS and an 8-bit bus interface below that
I’d add to that 64×32 (2k) icache at the top left – a TMS320?
how can you recognize all of this? O_o
(well I spent 10 years designing chips ….)
You can tell it’s 2-layer (or maybe 3) because the power rails are mixed in there and you can see them being routed under each other in places
You can count cells to get sizes (and guess since we’re mostly really only seeing one layer of metal here – some structures may be made from mirror imaged pairs and be twice the regular count you get)
some are guesses – I’ve actually had second thoughts about what I wrote above: the structure at the top seems to have tags and data – it’s an interesting structure though, might be fully associative, so maybe it’s a TLB instead – and looking more closely I see byte lanes in the data portion (to the right) so maybe it’s a data cache – being at the end of the data path is a more likely location for a data cache too
That kind of implies that the thing on the right is an instruction cache (but I see no tags unless the set size is large and that tiny block at the bottom is them) so maybe it’s an instruction ram for a DSP
In the data path (lower left) I can see a barrel shifter, I think the registers are the block just above half way, there’s an adder above it – the structure at the bottom looks like it might be a multiplier (I’m less sure about this one though – might be more likely a multiply/divide step)
To the right of the data path is instruction decode/data path control (being next to an icache/isram on the right makes sense)
This chip really is at a particular place/time – 2 layers of metal, but it has room for caches on board it’s sort of just after early Mips/Sparc timeframe
One more thing – there’s a 24-bit thing going on in the data path – might have 24-bit addressing – also the register decode seems to split the registers in half – maybe there’s two sets of 8 there
Maybe an early 68302? (is that a pcmcia on the right hand side?)
Hm, 24-bit and 32-bit mixed? PCMCIA?
Could this be an old Motorola MC68356?
Sorry, definitely a false guess – MC68356 seems to have much more pins.
Taniwha wrote:
> might have 24-bit addressing …
> seems to split the registers in half – maybe there’s two sets of 8 there
Both those layout features do sound 68K-ish.
Philips had a 68070 from around the right time period, but I don’t recall whether the available packages included a TQ144.
IIRC, the 68070 included some on-board peripherals and an MMU, I’ll see if I can dig up a datasheet.
-Brian
Awesome. Seems you were right about the TLB ;)
It has the pin count for an R3000.
I found a wonderful site of die photos that do not include this one at diephotos.blogspot.com. The NEC R3000 is there. I was wondering about i860 or i960, but found photos elsewhere that differ enough. My amateur eye sees just 2 layers of metal, and I don’t see mirroring of adjacent IO structures, so I lean towards suspecting that it is NMOS rather than CMOS (many CMOS structures are mirrored to share wells, but I’d expect it to be most obvious and differ from NMOS in the IOs).
I’m surprised that no one has yet mentioned i432, so I’ll start with that. Can’t find a die photo online yet – maybe Bunnie was able to arrange a global suppression?
Nevermind – searching for iAPX 432 was the trick, and it’s not a 432. National had the 32000 family, but for a while the largest package there was an 84 pin PGA, and a later 32532 had 175 pins, only 16 were reserved(NC?).
Maybe there is some vague resemblance to some of the AMD 29K family?
no not a 29k – we’d see a 192 entry register file if there was
And more contemporary chips:
http://www.computerhistory.org/microprocessors/up3.html
http://www.computerhistory.org/microprocessors/up4.html
But I don’t see a match.
Too many pins for the MC56000, not enough for the MC68356, I think.
I’m not sure the hidden message in the jpg helps – it’s obviously a 87?
Actually, the hidden message is longer than that. (Although the ware seems to be hard enough to analyze that I should probably reveal that the hidden message was just random mashing on the keyboard to entertain people who search for revealing metadata… :)
heh so I figured (or at least figured it was spurious) – I just posted it here as a little extra entertainment – it’s a great image – thanks
A SPARClite Fujitsu MB8683x ?