YC for Hardware

Sure, that's a fair statement. I'm not saying that the lack of success so far is a knock against YC for hardware startups. Rather, the similar lack of success for software startups in YC at the same stage of development, and the subsequent success they did enjoy, is not a point in favour of YC's suitability for hardware startups.

Hardware and Software startups are sufficiently different in their needs and life cycles that it would be specious to use the success of a particular program for the one as indicative of its suitability for the other. Similarly, the sorts of b2c startups that YC typically funds are sufficiently different from b2g startups like Palantir that I don't think you could claim Palantir would have benefitted from YC based solely on the experiences of b2c startups.

As others have noted, the pre-order route is often the best go to market strategy for new hardware companies, even if they have already raised a round. The buzz, evangelist community they build, feedback they receive, and the social proof they gather is often of a more lasting benefit than the pure pre-sales revenue. As others have already noted, Tilt has powered many of the recent massive hardware product launches that are going live on their own sites. More examples on https://open.tilt.com/preorders

Agreed. That is a good one.

I was thinking of more complex board, with non-standard finish and soldermask.

Even Pentalogix in Oregon offers 2-4 layer boards for about $25-$40 in two or three day turns.

NOTE: I am the founder of REX Computing (check my submission & comment history for more info), and thus a bit biased for this.

The standard thing I have heard from VC's that don't know anything about semiconductors (Or have not been involved in it in this century): "It will cost $100 million and take 3 years to have your first silicon. You have a 50% chance of even making it that far, and even then you only have a 10% chance of success (e.g. making any sales)"

So basically, the VC assumption is that it is a minimum $100 million and 3 years to see ANYTHING come out of it... and even if you get to silicon, there is only a 10% chance you'll be able to sell it. This is ridiculously pessimistic if you actually have a real design innovation.

Some of the new developments: New tools: High Level Synthesis an "Hardware construction languages" (such as Chisel... http://chisel.eecs.berkeley.edu) radically bring RTL development time down. My team (consisting of just two of us) were able to complete the RTL of our new processor core in under 4 months (working on the weekends)... and that includes a lot of redesign and working on things other than the RTL itself. Chisel also brings the testing/verification time down significantly, as it can auto-generate Verilog optimized for either VLSI or FPGAs, and a high speed cycle accurate C++ simulator of the chip that is ~10x faster than the industry standard Verilog C Simulator. Even though Chisel really is this amazing, virtually no one outside of UC Berkeley knows about it.

2. FPGAs have gotten a lot better... I can say the synthesis tools still suck, but in terms of speed and testing ablility, you can test your chip on very cheap (Ours is $300) FPGAs these days.

3.Shuttle run/Multi project wafer services... a "block" (6 mm^2) on TSMC 28nm process is $90,000... and you get 100 chips back. A 16 core version of our chip is just two of these blocks.

4. Older process nodes are getting cheaper. Most fabs in the US are not going below 65nm or 45nm, and need to keep full utilization of their fabs. There are a lot of deals that can be had on these smaller fabs which are perfect for prototyping... A similar block size can be as cheap as $3 to 5k.

5. Moore's law dying is one of the best things that can happen to the industry. Instead of the ridiculous investment getting to the next process node, the fabs can invest in making the current process better and cheaper. This will bring down the cost of very good, proven processes (like 28nm, and potentially 14/16nm in ~3 to 5 years), making fabrication even cheaper and more widely available.

I'm happy to answer any questions!

EDIT/TLDR: Based on my own startup's current plan and estimates, it is possible to prototype (get 100 chips back) an entirely new processor architecture for under $2 million, and it is possible to go into mass production with that design for $10 to $15 million.