HackerNews中文版

DNA 数据存储的宣传通常以密度和寿命为切入点。然而，一家位于伯克利的深度科技公司 BioCompute 将其定位为每字节成本问题，因为这是整个领域生死攸关的数字。该数字之所以停滞不前，是结构性的原因。几乎所有人都通过合成 DNA 来写入数据，逐个碱基构建新的链。当你的写入步骤是制造时，你的成本曲线就是合成的成本曲线，而且它下降得很慢。更糟糕的是，每条链都是一次性使用的，所以介质是消耗品。下次写入时，你又要支付制造费用。 BioCompute 的赌注是，如果你停止制造，你将获得一条不同的曲线。它通过使用酶标记可重复使用的模板来写入数据，而不是构建链，并通过纳米孔读取它们。模板不会被消耗。将合成从关键路径中移除，主要成本就会转移到你可以真正着力的地方。到目前为止，该公司报告的已证明的写入成本为每兆字节 1 美元。其目标是每太字节 1 美元。前者是团队已经展示的。后者是重复使用方法旨在达到的目标。值得仔细审查的部分是读写耦合。大多数团队优化一端，然后在另一端附加现有的测序技术。BioCompute 正在调整其授权的纳米孔读取过程以适应其自身的写入化学，因此两者是相互设计的，而不是简单地组合在一起。如果成本要降低几个数量级，这种契合度将是其中很大一部分的来源。对于任何关注过这个领域过度承诺的人来说，以下是诚实的开放性问题： * 已证明的成本是在小规模下实现的，并且不能保证随着产量的增加，经济效益能够维持。 * 大规模读取的准确性尚未得到证实，这在所有方面都是如此。 * 写入成本仍然有一个主要驱动因素，而整个目标取决于降低它。 BioCompute 有足够的实力值得认真对待。伯克利和斯坦福大学的资深学者为该公司提供咨询，该公司已为其写入方法申请了专利，并与两家美国创意工作室进行了早期付费试点。摆在桌面上的主张是狭窄且可检验的：重复使用加上协同设计的读写堆栈可以以合成永远无法做到的方式降低每字节的成本。这才是值得关注的。

查看原文

The pitch for DNA data storage usually leads with density and longevity. BioCompute, a deep-tech company in Berkeley, however, frames this as a cost-per-byte problem, because that is the number the whole field lives or dies on.The reason that number has been stuck is structural. Almost everyone writes data by synthesizing DNA, building new strands base by base. When your write step is manufacturing, your cost curve is the cost curve of synthesis, and it bends slowly. Worse, every strand is single-use, so the medium is a consumable. You pay the manufacturing cost again on the next write.BioCompute's bet is that you get a different curve if you stop manufacturing. It writes data by marking reusable templates with an enzyme instead of building strands, and it reads them back with a nanopore. The template is not consumed. Take synthesis off the critical path and the dominant cost moves somewhere you can actually push on.So far the company reports a demonstrated write cost of $1 per megabyte. Its target is $1 per terabyte. The first is what the team has shown. The second is where the reuse approach is built to drive it.The part worth scrutinizing is the read and write coupling. Most groups optimize one side and bolt on whatever sequencing exists for the other. BioCompute is tuning its licensed nanopore read process to fit its own write chemistry, so the two are engineered against each other rather than glued together. If the cost is going to fall by orders of magnitude, that fit is where a lot of it has to come from.The honest open questions, for anyone who has watched this field overpromise:- The demonstrated cost is at small scale, and there is no guarantee the economics survive as volume climbs.- Read accuracy at scale is unproven, which is true across the board.- The write cost still has one dominant driver, and the whole target rides on bringing it down.BioCompute has enough behind it to take seriously. Senior academics at Berkeley and Stanford advise the company, it has filed on its write method, and it has run early paid pilots with two US creative studios. The claim on the table is narrow and testable: reuse plus a co-designed read and write stack can move the cost per byte in a way that synthesis never will. That is the thing to watch.

BioCompute 致力于实现一个目标：让一美元能够购买到一百万太字节的存储空间。