Hacker News Reader

Comment by londons_explore | Hacker News Reader

So the intended market for these are...

Lab gear...?

Does it make sense in anything with higher sales volumes, or would it always make sense to make dedicated silicon with hardware offload for one specific protocol?

5G/6G base stations?

View on HN

The primary use for these is implementing Software-Defined Radios and the primary user of these is the military. SDRs allow jamming resistance, operation in hostile radio zones, and flexible network topologies.

These don't make sense for high volume. They are competing with ADI data converters + FPGA combo on PCBs, which are expensive. Comparable ADI ADCs are 10k+ usd/SKU. Unfortunately for these applications cost and compactness might not be the biggest priority. ADI has great customer support for example, which is really useful while developing such an high end product.

Funny thing is that these ADC/DAC/RF front-ends aren't ghat uncommon. A lot of wireline or wireless SoCs have such high performing blocks, but they aren't general purpose. They are rather optimized for the application and the performance far exceeds this kind of general purpose stuff.

Oh, also, these days one can buy IPs from companies like Socionext or OmniDesign which are in a similar space. These have IPs in 16/7/5nm typically. Anyone willing to spend 30 milion+ usd can have an SDR chip in volume production in like 4-5 years, which is a typical development cycle.

I have no experience with the last 2 companies you mentioned. Do they have the kind of eFPGA IP I've seen sometimes or do they only make CPU or fixed logic designs? The FPGA in the rfsoc has a lot of logic and DSP resources. It'd be a shame to give that up.

They do deliver the RF part, the customer design the DSP themselves, often by just writing the RTL but letting a 3rd party ASIC design services company to do the physical design. eFPGAs aren't popular, but for a good reason. One can add as much as programmability to a custom designed ASIC as possible, but specific to the application. It's extremely rare that one needs their own ASIC with general purpose logic. If you know your use case good, you don't need that.

Custom designed DSP almost always outperforms FPGAs because you can get the most out of the process technology. Also, FPGAs are often in 2-3 generation old process nodes. It is just way more expensive, but for these customers the performance edge often outperforms the price. High frequency trading companies cone to mind. They have transitioned from FPGAs to ASICs wherever it makes sense.

5G basestations are on dedicated silicon already.

TEMPEST spy stuff too maybe?

Reply on news.ycombinator.com