There is some question as to if the KNC Solar ever actually existed outside of a test chip IF at all - and there is ZERO doubt it was not in actual PRODUCTION in 2015 and probably not in early-to-mid 2016 as KNC hashrates didn't start climbing 'till somewhere in 2016 to any significant degree that would indicate deployment of miners based on a new and more-efficient chip.
The S5 and Spondoolies SP20 (and it's same-chip siblings like the SP30) had almost identical efficiency (give or take manufacturing variations) if you tuned the SP20 to near it's mid-point on mining chip operating voltage, but the SP20 WAS tuneable for higher hashrates at the cost of lower efficiency, or could be tuned for better efficiency at the cost of lower hashrate, when compared to the S5.
Those fall a LITTLE out of the "2 year" window on the basis of "when introduced" though most were still in use in that window.
BW.COM also announced *3* different chips and had announced plans to sell a miner on the basis of the first chap, but they appear to have decided to keep those chips and miners in-house.
The initial BW.com BW1401 (?) chip was specified as almost identical efficiency to the 28nm BM 1385 despite being on a 14/16nm process node - the successor chips were specified to be more efficient, with the third announced chip in the same range as the BM 1387/BF16/Avalon 741 type chips.
The S9 and R4 today are the most efficient miners (depending on the specific batch and to some degree specific machine variations) with the T9 and Avalon 741 / 721 close behind, out of miners that HAVE been evaluated in significant quantities so far.
Both the S7 and the S9 were the most efficient miners on the market for a period of months after they first started shipping. Both were challenged but not quite matched by Avalon units, the Avalon 6 IIRC came out only a few months after the S7, while the Avalon 721 lagged the S9 by quite a bit (the 741 was introduced a few months AFTER the 721).
Bitfury is a bit up in the air, some BifFury machines have been announced but seem to be vary scarce so far in the wild and have somewhat varied efficiencies announced.
It appears they may have had to refine the design before they actually got it into production, *OR* more likely they were able to use all of the chips they could get made internally for a while and didn't make any available to others (possible exception for long-time major-farm partners like MBP) because they didn't have any to SPARE until recently.
Innosilicon at one point announced an A3 chip for Bitcoin mining, but that appears to have never made it to production.
Based on actual real world performance of the A4 for Scrypt (announced at the same time as the A3 was announced), it would appear that the A4 was going to be similar efficiency to the BW 1401, and the introduction of the S7 caused Innosilicon to decide to cancel the A3 as a production chip as it would have been similar efficiency but a lot more expen$ive to make.
I do have to wonder if Innosilicon is still working on a "this generation" SHA256 mining chip to compete with the BM1387.