The miniaturization of semiconductor transistors has driven the growth in computer performance for more than 50 years. As miniaturization approaches its limits, bringing an end to Moore’s law, performance gains will need to come from software, algorithms, and hardware. We refer to these technologies as the “Top” of the computing stack to distinguish them from the traditional technologies at the “Bottom”: semiconductor physics and silicon-fabrication technology. In the post-Moore era, the Top will provide substantial performance gains, but these gains will be opportunistic, uneven, and sporadic, and they will suffer from the law of diminishing returns. Big system components offer a promising context for tackling the challenges of working at the Top.

No abstract

Deep learning's recent history has been one of achievement: from triumphing over humans in the game of Go to world-leading performance in image recognition, voice recognition, translation, and other tasks. But this progress has come with a voracious appetite for computing power. This article reports on the computational demands of Deep Learning applications in five prominent application areas and shows that progress in all five is strongly reliant on increases in computing power. Extrapolating forward this reliance reveals that progress along current lines is rapidly becoming economically, technically, and environmentally unsustainable. Thus, continued progress in these applications will require dramatically more computationally-efficient methods, which will either have to come from changes to deep learning or from moving to other machine learning methods.

No abstract

No abstract

Gene synthesis enables creation and modification of genetic sequences at an unprecedented pace, offering enormous potential for new biological functionality but also increasing the need for biosurveillance. In this paper, we introduce a bioinformatics technique for determining whether a gene is natural or synthetic based solely on nucleotide sequence. This technique, grounded in codon theory and machine learning, can correctly classify genes with 97.7% accuracy on a novel data set. We then classify ∼19,000 unique genes from the Addgene non-profit plasmid repository to investigate whether natural and synthetic genes have differential use in heterologous expression. Phylogenetic analysis of distance between source and expression organisms reveals that researchers are using synthesis to source genes from more genetically-distant organisms, particularly for longer genes. We provide empirical evidence that gene synthesis is leading biologists to sample more broadly across the diversity of life, and we provide a foundational tool for the biosurveillance community.

Competition between firms to invent and patent an idea, or “patent racing,” has been much discussed in theory, but seldom analyzed empirically and never at scale. This article introduces an empirical way to identify patent races, and provides the first broad-based view of them in the real world. It reveals that patent races are common, particularly in information-technology fields. The article then analyzes the effect of winning a patent race, showing that patent race winners do significantly more follow-on innovation, and their follow-on research is more similar to what was covered by the patent. (JEL CODES: O34, O32, O31)

No abstract