What's inside the cheapest and fakest jellybean opamps

[magic]

That's a question I asked a few months ago after getting some fake opamps myself and seeing countless reports of even the most mundane ICs being counterfeited.

Courtesy of AliExpress and their coupons I obtained a number of too cheap to be true opamps for virtually nothing straight from the source For a good measure I also bought samples from local resellers whenever I saw suspiciously looking chips at a place I was buying anything else from. Sure enough, wherever you get them from, it's all the same junk.

I wondered whether to test them or tear them down. Since a few types I had in quantities of 10, I decided to take the "don't turn it on" route with those. What I found inside quickly convinced me that testing will be utterly boring and instant teardown is the way to go.

Now a short description of the specimens and some pr0n. Image quality isn't quite Zeptobars level but good enough for identification

NE5532
I got a few to see if anything has changed since Zeptobars photographed one. Some turned out to be the same, but one was different.
Dice: 3, 659.

NE5534
As far as I know, there are no fully pin-compatible substitutes for this IC. So I wondered what could they put inside? Maybe there is a Chinese implementation of this classic design? Well, maybe not. They just don't care.
Dice: 8, ANON.

LM833
Another common audio opamp. Same as one of the 5532s, but bottom of the packages is different. The die looks quite funny, like a cartoon drawing.
Die: 659.

NJM4558
This one is popular with guitarists for whatever reason. Can we get the genuine JRC sound from AliExpress?
I actually had two, but destroyed the other one before taking the pic. It looked much more legit than this one, only the lettering was laser etched rather than printed.
Dice: 3, 46334.

TL082
Many cases reported of these behaving like LM358. What could be the reason?
Die: GI

TL071
What might be inside a single version of the above? More of the same, of course.
Die: 8, 20GI

LM358
If TL082 is LM358 than what is LM358?
Die: ADI6

UA741
Surely no one is going to fake the old 741, right?
Die: 20GI

OP07
This chip I actually bought long ago expecting to get a real one. Oh well.
Die: GI956

All in all, there isn't much rhyme or reason to it. Same part may have different dice, same die may be found in different parts, even BJT and JFET at the same time. Single opamps are realized by connecting one half of a dual die. As far as I can see, all dice implement LM358 or RC4558 topology. I will leave it as an exercise for the reader to figure out which are which

It could perhaps be interesting to test whether they meet the spec of LM358/RC4458, and they probably do. I suspect these are off-the-shelf products that fraudsters order without markings or as bare dice and print whatever BS on them. At scale it's probably cheaper than grinding markings off of recycled chips.

[Prehistoricman]

GI and 20GI are interesting. They look like they could have the same netlist, just laid out slightly differently.

What's with all the round vias in these chips? I thought that they were always square.

[amyk]

956 and 659 feel too close to be a coincidence... does anyone recognise the logo in the middle of GI956?

[schmitt trigger]

My theory is that counterfeiters will package whatever surplus or rejected dice they can source, and then label the plastic package with whatever part number remotely resembles the function.

After all, an opamp is an opamp, right ? 

[magic]

GI and 20GI are interesting. They look like they could have the same netlist, just laid out slightly differently.

As far as the signal path is concerned, all LM358 here implement the attached circuit.
This is from Fairchild LM358 datasheet, a version they got from the acquisition of Samsung in 1990s.

What's with all the round vias in these chips? I thought that they were always square.

You probably mean lateral PNP transistors
They can be any shape, though round is optimal because it provides uniform base width in all directions.

There are no vias here and the closest thing are "contact windows" - holes etched is the insulation layer between silicon and metal, permitting a connection. The location of those is clearly visible on ADI6, because metal appears brighter in those areas for some reason.

[magic]

All those Chinese LM358 appear to use identical bias generation circuit too. I don't understand how it works or what it is trying to achieve with all those transistors, but there it is.


I managed to cleanly etch one of the fake NE5534 chips and expose the wire bonding job inside. As could be expected, they use the left channel and run a wire across the whole die to connect its output to a pin on the right side of the package. The die is mounted slightly off-center, perhaps to prevent that wire from touching the VCC wire. Where there is a will, there is a way


Still waiting to encounter a dedicated half-LM358 die designed just for this purpose

[magic]

Still waiting to encounter a dedicated half-LM358 die designed just for this purpose

I learned that there actually exist single LM358/LM324 opamps such as this TS321 made by STM.
Why counterfeiters aren't using such dice is a mystery, perhaps nobody makes them cheaply enough due to lower production volume.

I had another look at the bias generator, found and corrected a certain mistake and figured out how it works.

This doesn't look much different from the previous version, but Q1 and Q2 became "NPN 3" and "NPN 2". This is LTspice's way of saying that they are 2 or 3 instances of the default NPN transistor connected in parallel. Indeed, in each of those opamps, Q2 always has twice the emitter area (= effective transistor size) as the smaller ones and is therefore equivalent to the parallel combination of Q4+Q5, and Q1 appears to be larger still.

So how it works? JFET current turns on Q4+Q5 and Q2, Q1 and the emitter followers Q4+Q5 prevent Q2 from saturating. This turns on Q3, which sinks majority of Q1 current, but leaves Q2 turned on because otherwise Q1 would slide down and turn off Q4+Q5.

The end result is two distinct current flows: almost all JFET current flows through Q1 and Q3 to ground, almost all Q4+Q5 current flows through Q2 and R1 to ground. There is a bit of base currents diverting here and there, but they are much smaller.

Let's calculate voltage at the top of R1. This is Vbe(Q3)+Vbe(Q4)-Vbe(Q1)-Vbe(Q2). Since Q2 and Q4+Q5 are effectively identical transistors running at equal currents, their Vbe are equal. Therefore R1 voltage is simply Vbe(Q3)-Vbe(Q1). Since Q1 and Q3 also run at equal current and Q1 is larger, its Vbe is lower and the difference of Vbe in such case is proportional to absolute temperature ("PTAT") - a known fact.

This PTAT voltage across R1 generates a PTAT current in R1 which is cascoded by Q2 and split equally to the two channels of the opamp by Q4 and Q5. Bingo, the amplifier is now biased by a PTAT current which cancels the usual transconductance droop of BJTs due to increasing temperature and maintains constant open loop gain and bandwidth.

The TI version also uses a constant transconductance / PTAT biasing circuit, but a more traditional one employing two NPNs and two PNPs, like found in LM317 or NE5532. It also has some additional diodes for startup, which are not necessary in the Chinese version.

The Chinese circuit shows some dependence on JFET current but not a lot. Below, I replaced the JFET with a current source stepped from 10µA to 100µA. The output is a straight line pointing close to absolute zero in each case.


Finally, I call it "Chinese circuit" but I have no idea who came up with it. Perhaps it's a well known bias generator and I'm just showing my ignorance But I haven't seen it before...

[Sacodepatatas]

Hello,

I've got some cheap counterfeit opamps from Aliexpress and I tore down some of them. I found that inside the Njm4558d in Sop8 footprint there is a dice number 3 like the one that is shown at the first post. However I found that the Njm4558d in DIP-8 and the NE5532 in SOP-8 share the same dice but I don't have any clue of what architecture implements. Do you know if this is a variant of 3? When tested in a single supply setup as a voltage follower, everyone exhibit common mode voltage imput inversion at around 1.2V.

What do you think? A new and improved design for a generic chinese dual opamp? Dice size is squarer than of dice 3, but a bit smaller.

[magic]

Hi.

Phase reversal at the negative rail is hallmark of RC4558, TL072 and similar basic two stage designs with a simple LTP input stage. It's pretty much inevitable in this topology.

So it seems safe bet yours is yet another 4558 variant.

BTW, I think I have never said which chip is which in the end? Well, it seems no one cared to attempt analyzing them so I will just give the answers: the first three are 4558-style, including die 3, everything else is LM358 clones.

Your die also doesn't really have enough transistors for a 358 but it very likely fits 4558, as show below:
red: input PNP emitters (shorted together) and collectors; observe that IN- collector goes only to the mirror but IN+ goes further down
orange: the NPN current mirror
yellow: likely the bias current mirror PNPs; one branch seems to feed the input stage
green: the VAS and output stage bias spreader have to be somewhere here
cyan: output NPN
blue: output PNP
violet X: compensation cap
pink X: the cap at the current mirror

Simple

[Sacodepatatas]

Wooowww, what a brilliant explanation! Thank you so much.

The first IC that I decapped was from a lot of opamps labeled as LM833 but were DOA, because i was so intrigued about the IC internals. When i took some pictures of the die, i was wondering whether it was a dual opamp or not, because i see the symmetric patterns, but fail locating some of the mandatory elements in an opamp, such as the compensation capacitor.

Anyway, i have some genuine LM833 that work fine in my RIAA preamp.

[magic]

No idea what this last one is. It's absurdly tiny.

Perhaps some simple dual comparator

[Sacodepatatas]

No idea what this last one is. It's absurdly tiny.

Perhaps some simple dual comparator

Your guess is correct, i've put another fake lm833 on a breadboard, treated like it was a dual comparator with some pullup resistor at output, and it worked!! In fact it seems to be a basic, but nice comparator. It can't drain much current though. Vo(low) rises about 100mV/mA. Anyhow when Vcc=12V and R(pullup) = 1kOhm, Vo(low) is still below 1V. And it switches quickly. I've built two square wave oscillators (one per section) and they perform pretty decent. Rise and fall times within 200ns. I can get an aceptable square wave oscillating beyond 400kHz.

Why don't they mark these chips as genuine cheap comparators of their own brand? Whoever knows...

[magic]

Probably it was designed to be LM393/LM2903. It's entirely possible that you could buy more of those chips by ordering LM393 from Ali, but of course they could send you some other LM393 version or perhaps even LM358

I thought that the output voltage is rather lousy, but then I looked up National LM393 datasheet and it seems to match the spec of the original part.

I sometimes wonder if some of those ICs may be available as legitimate products from places like LCSC, with logo and part number of the real manufacturer and maybe even a datasheet. But I don't buy from LCSC so dunno.

[Jan Audio]

Why cant they copy it good ?
I mean, they throw some sand in a factory and chips come out, it wont cost extra sand.

[JohnnyMalaria]

Why cant they copy it good ?
I mean, they throw some sand in a factory and chips come out, it wont cost extra sand.

To make "good" ones, they'd need to test them and have some kind of quality control. I don't think that's part of their business model.

[Sacodepatatas]

More "asian amateur pr0n"...

Despite being so cheap (2.5 cents) die size is a bit bigger than "3". I haven't tested yet in a circuit. Seller is pending for feedback 🤣🤣🤣🤣.

[magic]

It's hard to see all the details, but based on that PNP with five collectors, with two of them going somewhere in the vicinity of what might be an NPN current mirror, I'm going to bet it's another LM358. Check if it works down to ground

Probably it has nasty crossover distortion too.

Why cant they copy it good ?
I mean, they throw some sand in a factory and chips come out, it wont cost extra sand.

It does take more sand to make a more complex circuit. And there are more parts to go wrong, rendering the whole unusable. Wherever you go, a 5532 always costs more than a 4558 so there is a perfectly sound business in "upgrading" the latter to the former

There is also sound business selling that stuff on auctions sites, because many buyers don't realize that 358 or 4558 is all they need It always gets me how much positive feedback the sellers of that junk receive.

[Sacodepatatas]

It's hard to see all the details, but based on that PNP with five collectors, with two of them going somewhere in the vicinity of what might be an NPN current mirror, I'm going to bet it's another LM358. Check if it works down to ground

Probably it has nasty crossover distortion too.

You're right again. Output goes down to 0V in voltage follower. And there is crossover as well. Not very noticeable at first sight, but easy to find in a 10x gain inverting amplifier.

I wonder if this manufacturer could make a 4558, why fitting a 358 die Inside instead.... Sounds silly.

But, it also may happen that a legit chip reference from a manufacturer results prone to be mistaken as another totally different chip, and, thinking of it, that could be intended.

Some years back i made a P6 preamplifier and i installed two Philips NE532D that i salvaged from an old CD drive. Because I knew that TI's NE5532s in SOIC are marked N5532, i thought the Philips version was missing a "5" digit. The sound quality that came out of this preamp was really awful.

It turns out that "NE532D" is how Philips call their LM358s 🤦🏻‍♂️🤦🏻‍♂️🤦🏻‍♂️🤦🏻‍♂️🤦🏻‍♂️

One thing that has come into my mind is that if I use the disguised 358s that i've got from Aliexpress into certain designs that need a 358 within (for example: a logic low close to 0V, single rail, etc), it would make hard to any copier for duplicating my designs because they won't work properly if the straight parts are used 😅

[Sacodepatatas]

Before I knew what was inside those fake LM833s, i was going to throw them away. Now that I know they are dual comparators, i'm being prone to buy more, so one month back i purchased 50 more pieces for a little bit more than 1€. I couldn't resist the temptation 😅. I'll have to check if these are the same as the ones i previously purchased.

[Terry Bites]

this

[magic]

Hopefully not just empty plastic chips

By the way, how do you decap them?

[Sacodepatatas]

I use the rosin method that i read at hackaday website. It turned out that i had colophony at home because i use it for making my soldering flux and also for making castillian soap.

I found it by a chance, because colophony is one of the ingredients for making castillian soap and is quite popular among the elders here in Spain. I went to a drugstore and asked for colophony, but the attendant looked at me like i was talking silly. Then i explained that colophony is the solid part of the pine resin, while the liquid part is called "turpentine spirit" ("esencia de trementina" in spanish). Then she said "ahhhh, you refer to turpentine stone", went to the storage room, and came back with a bag full of colophony chunks. I paid 3€ for half a kilogram and i think i'm having enough for centuries before i need some more.

The rosin method works fine for me. I process the SOP package in a whole, while i grind the DIP one to a small dice. I save the small emptied 5ml glass vitamin D bottles used for treating ostheoporosis. Then i use about 2 grams of rosin, a pencil torch, a small piece or copper wire, and a spare ceramic tile. I think that preparing the setup is more time consuming that the actual decap process. I put the IC buried into the rosin inside the bottle, and i heat the glass for about 4 minutes at open air and wearing a KN95 facemask. Then i pour the content on the tile. After that i take the solidified mixture and rinse it with a few ml of alcohol. The time i find the piece of rosin that holds the silicon die, i put it apart and then clean it with a few drops of acetone. I have a poor-man's microscope made of a lens taken from a cheap broken laser pointer.

I used the silicon die photos obtained by this manner for filing several claims on Aliexpress about counterfeit parts and i have been refunded most of times.

Btw, the LM833 that i've recently received aren't those fancy comparators, but the 659 opamps. Not so bad though.

[magic]

Interesting. I boiled ICs in colophony on an electric stove at circa 400°C and nothing happened except lots of smoke.
But it clearly works well for you. Maybe the trick is to reach very high temperature (like 500°C or more) before all the rosin burns out.

That lens is not even too bad, except for chromatic aberration. Maybe l.p. lenses aren't corrected for it because they don't need to. I used an HD webcam lens and a point and shoot camera with optical zoom for higher magnification. Canon A420, I think.

[Noopy]

Interesting. I boiled ICs in colophony on an electric stove at circa 400°C and nothing happened except lots of smoke.
But it clearly works well for you. Maybe the trick is to reach very high temperature (like 500°C or more) before all the rosin burns out.

At these temperatures the epoxy already decomposes. Perhaps this effect dominates the decapping process and the rosin is only kind of a "washing liquid" for the residues.   

[Sacodepatatas]

Interesting. I boiled ICs in colophony on an electric stove at circa 400°C and nothing happened except lots of smoke.
But it clearly works well for you. Maybe the trick is to reach very high temperature (like 500°C or more) before all the rosin burns out.

Yes, i noticed that i have much success if i heat the bottle as fast as i can, and, the less amount of rosin the better. The time i stop heating, there's half the amount of rosin left in the bottle. The only drawback is that the glass bottle is of one time use. If, once cool, i try to heat it up again, the bottle breaks down into shards.

At these temperatures the epoxy already decomposes. Perhaps this effect dominates the decapping process and the rosin is only kind of a "washing liquid" for the residues.   

Yes, i agree. I did an experiment. I heated some colofony until it became liquid. Then i dropped some blue HDPE snips from a milk bottle. The plastic vanished (dissolved) and the mixture was being homogeneous, but the moment i tried to dilute the rosin by using alcohol, the plastic appeared and collapsed into a kind of foamy material. This doesn't happen when i dissolve the mixture of rosin and epoxy from ICs, indicating that this may no longer be a polymer.