3 ms·

> As software engineers we're used to quick iteration cycles in everything. It might not seem to you that they are "quick", but compared to the physical world t

> As software engineers we're used to quick iteration cycles in everything. It might not seem to you that they are "quick", but compared to the physical world they are. That's also partly because (physical world) engineers don't invest the same intense effort to enable quick iteration cycles for their area as software engineers do. I cannot say whether this is a "culture problem" of engineering disciplines, or managers don't see a huge economic value that this would enable, and thus don't allow engineers to work on such topics. I at least somewhat know what I am talking about, since when I had to get used to some CAD program, I really felt set back by decades in comparison what are standard practices in software development. A little bit like the difference between programming in a modern programming language and programming in Excel. Thus, in some sense I am impressed by (e.g. mechanical) engineers who despite all this actually are capable to accomplish creating a product (and I do believe that if the practices improved, they could do so much more).

ein0p 2y ago
It's more fundamental than this, IMO. No matter how much you invest into faster iteration cycles, physical world just takes a long time, and the worst part is, you either have to get some very complicated thing right the first time, or add a sufficient buffer to your plans. Discovered a bug in your chip after initial tapeout? Add months to the schedule. Too much RF crosstalk in the PCB? Add a couple of weeks at least. Take Apple for example. Do you think they didn't invest into faster iterations? I'm sure they did. And each chip still takes 3+ years there, too. Qualcomm? About 3 years. Google? Also about 3 years (I heard of the upcoming 192GB TPU being developed over 2 years ago when I was there; it was announced this week). The "innovations" that they are talking about today were actually decided 3 years ago. 3 years ago was a very different time, of course. There's also another problem when it comes to chips. On the higher end, you have to design for the technological processes that don't yet exist. E.g. 2nm did not exist 3 years ago, yet the design of the chips that use it was done back then. You're also doing it with bleeding edge tooling.

ein0p 2y ago

It's more fundamental than this, IMO. No matter how much you invest into faster iteration cycles, physical world just takes a long time, and the worst part is, you either have to get some very complicated thing right the first time, or add a sufficient buffer to your plans. Discovered a bug in your chip after initial tapeout? Add months to the schedule. Too much RF crosstalk in the PCB? Add a couple of weeks at least. Take Apple for example. Do you think they didn't invest into faster iterations? I'm sure they did. And each chip still takes 3+ years there, too. Qualcomm? About 3 years. Google? Also about 3 years (I heard of the upcoming 192GB TPU being developed over 2 years ago when I was there; it was announced this week). The "innovations" that they are talking about today were actually decided 3 years ago. 3 years ago was a very different time, of course. There's also another problem when it comes to chips. On the higher end, you have to design for the technological processes that don't yet exist. E.g. 2nm did not exist 3 years ago, yet the design of the chips that use it was done back then. You're also doing it with bleeding edge tooling.