Information Efficient High-speed Signalling for Intelligent Computing

让我们面对现实吧--我们对与世界其他地方永久连接的渴望是昂贵的。与过去我们只是乐于与亲人分享简短的对话不同，现在我们通过分享包括视频、语音和文件在内的内容来开展业务。显然，如此大的数据是由每年传输和处理zettabytes数据的数据中心处理的。去年，数据中心消耗了416太瓦，未来看起来不妙。显然，当我们考虑碳足迹和冷却成本时，这是令人担忧的。联邦组织正在考虑采取严厉措施--2014年对匈牙利征收互联网税的讨论表明了形势的严重性。该措施可抵销冷却成本，但不会解决环境影响。因此，没有替代能源效率的互连技术，可以传输数据消耗更少的电力，也产生更少的热量。拟议的研究将探索实现这一目标的方法。第一，我们将通过允许它根据需要分配资源来建立“智能”联系。这需要重新考虑链路设计-采用基于ADC（模数转换器）的数字并行处理是必要的。其次，解决诸如符号间干扰（ISI）的信道损伤增加了功耗和复杂度。我们将建设性地使用ISI，这不仅简化了均衡过程，还使我们能够实现更低的BER。第三，我们需要找到一种智能的方式来发送数据位。传统上，使用信号的幅度来表示传输的数据。显然，发送多个比特需要更高的幅度，这是负担不起的。因此，我们需要探索其他调制技术。最后，链接必须是智能的，以便利用传输数据中的“内容”。总的来说，对这些想法的探索将继续在后摩尔定律时代提高速度和能效，以帮助设计未来的数据中心

Let's face it - our desire for perpetual connectivity to rest of the world is expensive. Unlike old days when we would be just happy to share a short conversation to our loved ones, these days we conduct business by sharing content including video, voice and files. Obviously, such large data is handled by data centers that transfers and process zettabytes of data every year. Last year, data centers consumed 416 Terawatts and future looks ominous. Obviously, it is alarming when we consider carbon footprint and cooling cost. Federal organizations are considering drastic measures - discussion of internet taxing Hungary in 2014 indicates the gravity of the situation. Such measure may offset the cooling cost but would not address the environment impact. Therefore, there is no substitute to energy efficient interconnect technology that can transfer data consuming lot less electricity and also generates less heat. The proposed research will be exploring ways to enable that. First, we will make the links for `intelligent' by allowing it to allocate resources according to the need. This requires rethinking the link design - adoption of ADC (analog to digital converter) based digital parallel processing is necessary. Second, addressing channel impairments such as inter symbol interference (ISI) increases power consumption and complexity. We will constructively use the ISI that not only simplifies the equalization process, it also allows us to achieve lower BER. Third, we need to find an intelligent way to send the data bits. Traditionally, amplitude of a signal was used represents transmitted data. Obviously, sending multiple bits demands higher amplitude that is not affordable. Therefore, we need to explore other modulation techniques. Finally, links need to be intelligent to take advantage of the `content' in the transmitted data. Overall, exploration of these ideas will continue speed and energy efficiency improvement in post Moore's law era to help design future data centers