The University of Michigan’s new Center for Applications Driving Architectures (ADA) aims to make ASIC development easier and more affordable

This is a laudable effort, considering that the cost of developing a non-trivial (i.e., much more complex than the dumb ASICs for bitcoin mining) runs in the tens of millions of dollars.  However, I predict that most firms will still start their development with FPGA.  Of course, FPGA has much lower clock frequency and is not as energy-efficient as ASIC, its cost is orders of magnitude lower.

https://news.engin.umich.edu/2018/01/reimagining-how-computers-are-designed-university-of-michigan-leads-new-32m-center/

http://blog.zorinaq.com/asic-development-costs-are-lower-than-you-think/

http://blog.ebv.com/asic-expensive-part/

http://anysilicon.com/asic-right-next-iot-product/

https://toshiba.semicon-storage.com/eu/product/custom-soc/platform/ffsa.html

Timescale: a SQL and ACID-compliant open-source database for time-series data that is claimed to be faster than NoSQL databases

You don’t run into this everyday: a database that uses SQL natively, meets ACID requirements, handles time-series data, open-source, and yet competitive (or so claimed) against NoSQL databases.  If I were still in the high-frequency trading business, I would definitely take a look.

The best part is its use of SQL.  No one really wants to deal with MapReduce or NoSQL.  At the end of the day, if you are honest with yourself, you will admit that you want to use SQL to get stuff done.  We use NoSQL only because of perceived performance advantage.  No one wants eventual consistency if immediate consistency is available at a small performance price.

http://www.timescale.com/

https://blog.timescale.com/

The real innovation is happening among the very few companies that build their own deep-learning frameworks from scratch, both for training and multi-device inference.

…the publicly available frameworks like TensorFlow are great for research but are not efficient enough for deployment in cybersecurity….  using convolutional neural nets is good with local correlations in the data [but not so for situations without local correlation]….

https://www.nextplatform.com/2018/01/26/startup-builds-gpu-native-custom-neural-network-framework/

A WSJ Reporter’s 10-Year Odyssey Through America’s Housing Crisis

This is a great personal story and facts about the mortgage crisis that precipitated the financial crisis of 2007 (or 2008, depending on whom you ask), from the eyes of a Wall Street Journal reporter who owed more money on his mortgage than the house was worth.

https://www.wsj.com/articles/my-10-year-odyssey-through-americas-housing-crisis-1516981725

Google’s lack of success in consumer products; the Grab revolution in Southeast Asia has already happened in China.

This is an interesting post, and I like the author’s writing style.

I think his experience at Google basically shows that Google is no good at consumer products. All of the winning products the author lists (e.g., Spanner, TPU) are enterprise products, and all of the failing ones are consumer products. Unfortunately, Google Cloud, arguably their most important enterprise product, is faltering and falling farther and farther behind AWS.

Another point I wanted to make is that the author would be advisable to look to China for the next revolution, where he will see a lot of parallels to his experience in Southeast Asia. WeChatPay is already widely used in China, as most Chinese are already going cashless. The revolution has already happened in China.

https://medium.com/@steve.yegge/why-i-left-google-to-join-grab-86dfffc0be84