18 months ago Renesas announced they were prototyping their SG-MONOS eFlash on 28nm, and at the time we said it would be a couple of years before actual product. Yesterday, Renesas revealed their partner in this effort is TSMC – no surprise – and hinted things are moving, with better performance than expected but on a longer qualification timeline than anticipated.
TSMC and Renesas have been working together on eFlash for MCUs since their 90nm efforts. Four years ago, the pair completed a 40nm implementation. Bringing together the Renesas Metal-Oxide-Nitride-Oxide-Silicon (MONOS) eFlash with TSMC’s 28nm high-K metal gate process delivers temperature range support suitable for automotive MCUs.
From previously released information, Renesas SG-MONOS is an attempt to solve four problems with reduced eFlash geometries:
1) High-speed readout with high reliability: Decreased cell current on smaller memory cells could be solved with overdriving the word line, but that would push reliability in the wrong direction at high temperature. Adding a negative temperature dependence reduces the amount of overdrive and improves transistor wear by a factor of 10.
2) Mitigating erase voltage stress: Again, high temperature issues appear with the interlayer dielectric. By monitoring erase speeds and suppressing the maximum erase voltage during high-speed erase, the stress is reduced and reliability improves.
3) Increasing write speeds: Write pulse durations have been reduced using negative back bias on the memory cell, and writes to multiple eFlash cells are done in parallel.
4) Reducing power supply noise and EMI: Spread-spectrum clock generation on the drive clock for the charge pumps generating voltage for eFlash rewrites cuts the noise significantly.
In other words, these aren’t strictly process issues – most of these fixes seeking better reliability are in the eFlash programming and controller circuitry. Renesas didn’t provide an update on what other steps they are taking with TSMC to get to the levels of reliability needed for extended temperature range automotive MCUs. We can assume the challenges are substantial; the announcement says TSMC 28nm eFlash MCU samples are expected in 2017, with full production in 2020, which would be eight years after the previous 40nm node went to production.
Once reliability goals are realized, 28nm increases program memory capacity by 4x over 40nm, and by the latest indications performance also increases 4x, more than Renesas initially thought. Along with the eFlash enhancements for next-generation automotive MCUs, Renesas also expects to deliver multicore technology, better security features, and advanced interfaces.
The joint press release spends a lot of time talking about the benefits of flash-enabled MCUs in automotive applications, which many bloggers have discussed in SemiWiki, but the big one may be better reliability for what is expected to be more over-the-air (OTA) updates.
One unspoken truth in all this is how carefully the pair is proceeding to ensure reliability. With the MCU count per car continuing to rise, automotive manufacturers can ill afford service upticks due to failing MCUs. Liability, particularly as autonomous-driving looms, also is a huge concern. This effort has turned out to be a lot more than a mere process shrink, and would indicate a bump beyond 28nm might require a shift in embedded flash technology for automotive temperature grade parts.
By that same token, this level of investment to get to 28nm eFlash should give Renesas an advantage in automotive MCUs when they are ready. For those so inclined to read the science, a Renesas team published a detailed paper on the 28nm SG-MONOS eFlash technology in the IEEE Journal of Solid State Circuits earlier this year.