在单个调制中展示四象限模拟内存矩阵乘法
Manuel Le Gallo1, Oscar Hrynkevych1,2, Benedikt Kersting1
1IBM Research Europe, 8803 Rüschlikon, Switzerland.
Npj unconventional computing
|October 7, 2024
概括
本研究介绍了一种用于模拟内存计算 (AIMC) 的新型四象限矩阵向量乘法 (MVM). 这种新方法提高了人工智能任务相变内存芯片的准确性和效率.
科学领域:
- 材料科学 材料科学 材料科学
- 计算机工程 计算机工程
- 人工智能的人工智能
背景情况:
- 模拟内存计算 (AIMC) 使用电阻式内存设备进行诸如矩阵向量乘法 (MVM) 等计算.
- 单相MVM可以减少延迟,但由于设备导电的电压极性依赖,导致准确性问题.
研究的目的:
- 开发一个4象限MVM技术,减轻AIMC中的导电极性依赖.
- 在多核相变内存 (PCM) 芯片上实现和验证此技术.
主要方法:
- 开发了模拟和数字校准程序,以解决导电极性的依赖性.
- 在单一调制方案中实施了4象限MVM.
- 使用基于相位变换内存的多核AIMC芯片.
主要成果:
- 成功展示了精确的神经网络推断和相似性搜索任务.
- 与传统的四相阅读方案相比,实现了4倍以上的MVM吞吐量.
- 在AIMC运营中显著提高了能源效率.
结论:
- 开发的校准方法有效地减轻了AIMC中的导电极性依赖.
- 四象限MVM方法可以在基于PCM的AIMC芯片上进行高效和准确的计算.
- 这项工作使AIMC在需要高吞吐量和能源效率的实际AI应用中取得了进展.
相关概念视频
01:20Design Example: Capacitance Multiplier Circuit
718
In integrated circuit technology, a capacitance multiplier is often utilized to produce a larger capacitance value when a small physical capacitance falls short. This is achieved by a circuit that multiplies capacitance values by a factor of up to 1000, such that a 10-pF capacitor can replicate the performance of a 100-nF capacitor.
The circuit illustrated in Figure 1 below incorporates two op-amps, with the first operating as a voltage follower and the second acting as an inverting amplifier.
The circuit illustrated in Figure 1 below incorporates two op-amps, with the first operating as a voltage follower and the second acting as an inverting amplifier.
718
01:13Phasor Arithmetics
251
Phasors and their corresponding sinusoids are interrelated, offering unique insights into the behavior of alternating current (AC) circuits. One way to understand this relationship is through the operations of differentiation and integration in both the time and phasor domains.
When the derivative of a sinusoid is taken in the time domain, it transforms into its corresponding phasor multiplied by j-omega (jω) in the phasor domain, where j is the imaginary unit, and ω is the angular...
When the derivative of a sinusoid is taken in the time domain, it transforms into its corresponding phasor multiplied by j-omega (jω) in the phasor domain, where j is the imaginary unit, and ω is the angular...
251
01:24Bus Impedance Matrix
110
Calculating subtransient fault currents for three-phase faults in an N-bus power system involves using the positive-sequence network. When a three-phase short circuit occurs at a specific bus, the analysis uses the superposition method to evaluate two separate circuits.
In the first circuit, all machine voltage sources are short-circuited, leaving only the prefault voltage source at the fault location. The positive-sequence bus impedance matrix can be determined by solving the nodal equations,...
In the first circuit, all machine voltage sources are short-circuited, leaving only the prefault voltage source at the fault location. The positive-sequence bus impedance matrix can be determined by solving the nodal equations,...
110
01:21Bulk Modulus
292
The bulk modulus is a scientific term used to describe a material's resistance to uniform compression. It is the proportionality constant that links a change in pressure to the resulting relative volume change.
292
01:22Sum and Difference OpAmps
672
Operational amplifiers (op-amps) are versatile devices that extend beyond amplification. In this context, two specific op-amp configurations are explored: the summing and difference amplifiers.
A summing amplifier, or an adder, utilizes an op-amp to merge multiple input signals into a single output signal. When audio signals are introduced into its input channels, the input resistors initiate currents that traverse feedback resistors, resulting in an output voltage. Applying Kirchhoff's...
A summing amplifier, or an adder, utilizes an op-amp to merge multiple input signals into a single output signal. When audio signals are introduced into its input channels, the input resistors initiate currents that traverse feedback resistors, resulting in an output voltage. Applying Kirchhoff's...
672
01:16Basic Discrete Time Signals
200
The unit step sequence is defined as 1 for zero and positive values of the integer n. This sequence can be graphically displayed using a set of eight sample points, showing a step function starting from n=0 and remaining constant thereafter.
The unit impulse or sample sequence is mathematically expressed as zero for all n values except at n=0, where it is one. The unit impulse sequence, denoted by δ(n), is the first difference of the unit step sequence, while the unit step sequence u(n) is...
The unit impulse or sample sequence is mathematically expressed as zero for all n values except at n=0, where it is one. The unit impulse sequence, denoted by δ(n), is the first difference of the unit step sequence, while the unit step sequence u(n) is...
200