Hi all,

I have basic knowledge in electronic engineering but feel a bit rusty since I haven't been doing this for a couple of years.

The project:

• I've created a PCB to program empty Atmega328P chipset, when connected to Arduino (I'm using ELEGOO UNO R3)
• It houses a TQFP32 to DIP32 IC test socket
• This setup should allow to flash the bootloader an upload Arduino code, all through the 6pin ICSP connector

The schematic shows the Atmega328P chip but on the PCB layout i'm using the adapter (i've made sure I've mapped the adaptor pins correctly)

I had the basic design but then used DeepSeek to perfect it - both the schematic and the PCB layout.

Can you guys tell me if this looks right to you? It would really help me understand if DeepSeek is actually useful to a hobbyist like me. Cheers

 

Why do you need AI for something as simple as a test socket? What is perfection to a program? From a very quick look, it seems your XTAL loading capacitors are placed wrong.

The correct way.


First actually learn the process of making something like this so, you can detect when these programs screw something complex and complicated up instead screwing up something simple like this.
No, I don't think BS-ing, hallucinating programs like DeepSeek help a beginning hobbyist to learn the art of PCB design.

 

As mentioned by NSAspook, the crystal oscillator won’t as drawn.
The crystal’s loading capacitors are a well known and ubiquitous circuit topology, that AI gets them wrong gives me hope that humans are not yet obsolete.

 

As mentioned by NSAspook, the crystal oscillator won’t as drawn.
The crystal’s loading capacitors are a well known and ubiquitous circuit topology, that AI gets them wrong gives me hope that humans are not yet obsolete.

AI today is a bad joke. these programs understand nothing, don't reason and can't be trusted to water a plant.

HIGH TECH SNAKE-OIL

 

it is still early, and like everything new it is still buggy and unreliable. ICSP pins are not even grouped together... but things will change over time...

 

it is still early, and like everything new it is still buggy and unreliable. ICSP pins are not even grouped together... but things will change over time...

I firmly believe we will get to AGI eventually but I also believe this current crop of LLM technology won't get us there. Very useful for exploring human interfaces in human language communication with machines but completely lacking in the essence of intelligence, reasoning and understanding.

 

Thank you guys,

The reason I picked a 'simple' design is to create it then post it here and then see how much DeepSeek actually messed up. From the looks of it, you only found one error so wouldn't you say that in the grand scheme of things, it did pretty well?

I didn't use best practises though. One should name all the nets in a 'simple' language for the AI to be able to see routing etc.

I have more complex designs. I really appreciate your input guys. Cheers

 

Thank you guys,

The reason I picked a 'simple' design is to create it then post it here and then see how much DeepSeek actually messed up. From the looks of it, you only found one error so wouldn't you say that in the grand scheme of things, it did pretty well?

I didn't use best practises though. One should name all the nets in a 'simple' language for the AI to be able to see routing etc.

I have more complex designs. I really appreciate your input guys. Cheers

No, it didn't do well. It screwed the pooch failed on something fundamental and easy. I wouldn't trust it to design a resistor divider. I'm sure you will be much better with a little more practice after actually learning about the subject.